]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/kdebug.c
xnu-2782.30.5.tar.gz
[apple/xnu.git] / bsd / kern / kdebug.c
index 286a9d8ce2387de0c7f9ee02db0b274b8b2d2a3e..65c98080d0554a3c64b094ddd464793dee2ccc07 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2013 Apple Inc. All rights reserved.
  *
  * @Apple_LICENSE_HEADER_START@
  * 
  * License for the specific language governing rights and limitations
  * under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 
+
 #include <machine/spl.h>
 
 #include <sys/errno.h>
@@ -30,6 +31,8 @@
 #include <sys/sysctl.h>
 #include <sys/kdebug.h>
 #include <sys/sysproto.h>
+#include <sys/bsdtask_info.h>
+#include <sys/random.h>
 
 #define HZ      100
 #include <mach/clock_types.h>
 #include <mach/mach_time.h>
 #include <machine/machine_routines.h>
 
+#if defined(__i386__) || defined(__x86_64__)
+#include <i386/rtclock_protos.h>
+#include <i386/mp.h>
+#include <i386/machine_routines.h>
+#endif
+
+#include <kern/clock.h>
+
 #include <kern/thread.h>
 #include <kern/task.h>
+#include <kern/debug.h>
+#include <kern/kalloc.h>
+#include <kern/cpu_data.h>
+#include <kern/assert.h>
+#include <kern/telemetry.h>
 #include <vm/vm_kern.h>
 #include <sys/lock.h>
 
 #include <sys/malloc.h>
+#include <sys/mcache.h>
 #include <sys/kauth.h>
 
+#include <sys/vnode.h>
+#include <sys/vnode_internal.h>
+#include <sys/fcntl.h>
+#include <sys/file_internal.h>
+#include <sys/ubc.h>
+#include <sys/param.h>                 /* for isset() */
+
 #include <mach/mach_host.h>            /* for host_info() */
 #include <libkern/OSAtomic.h>
 
+#include <machine/pal_routines.h>
+
+extern boolean_t kdebug_serial;
+#if KDEBUG_MOJO_TRACE
+#include <sys/kdebugevents.h>
+static void kdebug_serial_print(       /* forward */
+               uint32_t, uint32_t, uint64_t,
+               uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t);
+#endif
+
+/*
+ * IOP(s)
+ *
+ * https://coreoswiki.apple.com/wiki/pages/U6z3i0q9/Consistent_Logging_Implementers_Guide.html
+ *
+ * IOP(s) are auxiliary cores that want to participate in kdebug event logging.
+ * They are registered dynamically. Each is assigned a cpu_id at registration.
+ *
+ * NOTE: IOP trace events may not use the same clock hardware as "normal"
+ * cpus. There is an effort made to synchronize the IOP timebase with the
+ * AP, but it should be understood that there may be discrepancies.
+ *
+ * Once registered, an IOP is permanent, it cannot be unloaded/unregistered.
+ * The current implementation depends on this for thread safety.
+ *
+ * New registrations occur by allocating an kd_iop struct and assigning
+ * a provisional cpu_id of list_head->cpu_id + 1. Then a CAS to claim the
+ * list_head pointer resolves any races.
+ *
+ * You may safely walk the kd_iops list at any time, without holding locks.
+ *
+ * When allocating buffers, the current kd_iops head is captured. Any operations
+ * that depend on the buffer state (such as flushing IOP traces on reads,
+ * etc.) should use the captured list head. This will allow registrations to
+ * take place while trace is in use.
+ */
+
+typedef struct kd_iop {
+       kd_callback_t   callback;
+       uint32_t        cpu_id;
+       uint64_t        last_timestamp; /* Prevent timer rollback */
+       struct kd_iop*  next;
+} kd_iop_t;
+
+static kd_iop_t* kd_iops = NULL;
+
 /* XXX should have prototypes, but Mach does not provide one */
 void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
 int cpu_number(void);  /* XXX <machine/...> include path broken */
+void commpage_update_kdebug_enable(void); /* XXX sign */
 
 /* XXX should probably be static, but it's debugging code... */
-int kdbg_read(user_addr_t, size_t *);
+int kdbg_read(user_addr_t, size_t *, vnode_t, vfs_context_t);
 void kdbg_control_chud(int, void *);
 int kdbg_control(int *, u_int, user_addr_t, size_t *);
-int kdbg_getentropy (user_addr_t, size_t *, int);
-int kdbg_readmap(user_addr_t, size_t *);
+int kdbg_readcpumap(user_addr_t, size_t *);
+int kdbg_readcurcpumap(user_addr_t, size_t *);
+int kdbg_readthrmap(user_addr_t, size_t *, vnode_t, vfs_context_t);
+int kdbg_readcurthrmap(user_addr_t, size_t *);
 int kdbg_getreg(kd_regtype *);
 int kdbg_setreg(kd_regtype *);
 int kdbg_setrtcdec(kd_regtype *);
 int kdbg_setpidex(kd_regtype *);
 int kdbg_setpid(kd_regtype *);
-void kdbg_mapinit(void);
-int kdbg_reinit(void);
-int kdbg_bootstrap(void);
+void kdbg_thrmap_init(void);
+int kdbg_reinit(boolean_t);
+int kdbg_bootstrap(boolean_t);
+
+int kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, uint32_t* cpumap_size);
+kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, unsigned int *mapsize, unsigned int *mapcount);
 
-static int  create_buffers(void);
+static int kdbg_enable_typefilter(void);
+static int kdbg_disable_typefilter(void);
+
+static int create_buffers(boolean_t);
 static void delete_buffers(void);
 
-#ifdef ppc
-extern uint32_t maxDec;
-#endif
+extern void IOSleep(int);
 
 /* trace enable status */
 unsigned int kdebug_enable = 0;
 
-/* track timestamps for security server's entropy needs */
-uint64_t *               kd_entropy_buffer = 0;
-unsigned int      kd_entropy_bufsize = 0;
-unsigned int      kd_entropy_count  = 0;
-unsigned int      kd_entropy_indx   = 0;
-unsigned int      kd_entropy_buftomem = 0;
-
+/* A static buffer to record events prior to the start of regular logging */
+#define        KD_EARLY_BUFFER_MAX      64
+static kd_buf          kd_early_buffer[KD_EARLY_BUFFER_MAX];
+static int             kd_early_index = 0;
+static boolean_t       kd_early_overflow = FALSE;
 
 #define SLOW_NOLOG     0x01
 #define SLOW_CHECKS    0x02
-#define SLOW_ENTROPY   0x04
+#define SLOW_ENTROPY   0x04                    /* Obsolescent */
+#define SLOW_CHUD      0x08
 
-unsigned int kdebug_slowcheck=SLOW_NOLOG;
+#define EVENTS_PER_STORAGE_UNIT                2048
+#define MIN_STORAGE_UNITS_PER_CPU      4
 
-unsigned int kd_cpus;
+#define POINTER_FROM_KDS_PTR(x) (&kd_bufs[x.buffer_index].kdsb_addr[x.offset])
 
-struct kd_bufinfo {
-        kd_buf * kd_stop;
-        kd_buf * kd_bufptr;
-        kd_buf * kd_buffer;
-        kd_buf * kd_buflast;
-        kd_buf * kd_readlast;
-        int      kd_wrapped;   /* plus, the global flag KDBG_WRAPPED is set if one of the buffers has wrapped */
-        uint64_t kd_prev_timebase;
-        int     kd_pad[24];    /* pad out to 128 bytes so that no cache line is shared between CPUs */
+union kds_ptr {
+       struct {
+               uint32_t buffer_index:21;
+               uint16_t offset:11;
+       };
+       uint32_t raw;
+};
+
+struct kd_storage {
+       union   kds_ptr kds_next;
+       uint32_t kds_bufindx;
+       uint32_t kds_bufcnt;
+       uint32_t kds_readlast;
+       boolean_t kds_lostevents;
+       uint64_t  kds_timestamp;
 
+       kd_buf  kds_records[EVENTS_PER_STORAGE_UNIT];
 };
 
+#define MAX_BUFFER_SIZE                        (1024 * 1024 * 128)
+#define N_STORAGE_UNITS_PER_BUFFER     (MAX_BUFFER_SIZE / sizeof(struct kd_storage))
+
+struct kd_storage_buffers {
+       struct  kd_storage      *kdsb_addr;
+       uint32_t                kdsb_size;
+};
+
+#define KDS_PTR_NULL 0xffffffff
+struct kd_storage_buffers *kd_bufs = NULL;
+int    n_storage_units = 0;
+int    n_storage_buffers = 0;
+int    n_storage_threshold = 0;
+int    kds_waiter = 0;
+
+#pragma pack(0)
+struct kd_bufinfo {
+       union  kds_ptr kd_list_head;
+       union  kds_ptr kd_list_tail;
+       boolean_t kd_lostevents;
+       uint32_t _pad;
+       uint64_t kd_prev_timebase;
+       uint32_t num_bufs;
+} __attribute__(( aligned(MAX_CPU_CACHE_LINE_SIZE) ));
+
+struct kd_ctrl_page_t {
+       union kds_ptr kds_free_list;
+       uint32_t enabled        :1;
+       uint32_t _pad0          :31;
+       int                     kds_inuse_count;
+       uint32_t kdebug_flags;
+       uint32_t kdebug_slowcheck;
+       /*
+        * The number of kd_bufinfo structs allocated may not match the current
+        * number of active cpus. We capture the iops list head at initialization
+        * which we could use to calculate the number of cpus we allocated data for,
+        * unless it happens to be null. To avoid that case, we explicitly also
+        * capture a cpu count.
+        */
+       kd_iop_t* kdebug_iops;
+       uint32_t kdebug_cpus;
+} kd_ctrl_page = { .kds_free_list = {.raw = KDS_PTR_NULL}, .kdebug_slowcheck = SLOW_NOLOG };
+
+#pragma pack()
+
 struct kd_bufinfo *kdbip = NULL;
 
-#define KDCOPYBUF_COUNT        1024
+#define KDCOPYBUF_COUNT        8192
 #define KDCOPYBUF_SIZE (KDCOPYBUF_COUNT * sizeof(kd_buf))
 kd_buf *kdcopybuf = NULL;
 
+boolean_t kdlog_bg_trace = FALSE;
+boolean_t kdlog_bg_trace_running = FALSE;
+unsigned int bg_nkdbufs = 0;
 
-unsigned int nkdbufs = 8192;
-unsigned int kd_bufsize = 0;
-unsigned int kdebug_flags = 0;
+unsigned int nkdbufs = 0;
 unsigned int kdlog_beg=0;
 unsigned int kdlog_end=0;
 unsigned int kdlog_value1=0;
@@ -122,6 +252,8 @@ unsigned int kdlog_value2=0;
 unsigned int kdlog_value3=0;
 unsigned int kdlog_value4=0;
 
+static lck_spin_t * kdw_spin_lock;
+static lck_spin_t * kds_spin_lock;
 static lck_mtx_t  * kd_trace_mtx_sysctl;
 static lck_grp_t  * kd_trace_mtx_sysctl_grp;
 static lck_attr_t * kd_trace_mtx_sysctl_attr;
@@ -135,10 +267,12 @@ static lck_mtx_t        stackshot_subsys_mutex;
 void *stackshot_snapbuf = NULL;
 
 int
-stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t options, register_t *retval);
+stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset, int32_t *retval);
 
+int  
+stack_snapshot_from_kernel(pid_t pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytesTraced);
 extern void
-kdp_snapshot_preflight(int pid, void  *tracebuf, uint32_t tracebuf_size, uint32_t options);
+kdp_snapshot_preflight(int pid, void  *tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset);
 
 extern int
 kdp_stack_snapshot_geterror(void);
@@ -148,11 +282,33 @@ kdp_stack_snapshot_bytes_traced(void);
 kd_threadmap *kd_mapptr = 0;
 unsigned int kd_mapsize = 0;
 unsigned int kd_mapcount = 0;
-unsigned int kd_maptomem = 0;
+
+off_t  RAW_file_offset = 0;
+int    RAW_file_written = 0;
+
+#define        RAW_FLUSH_SIZE  (2 * 1024 * 1024)
 
 pid_t global_state_pid = -1;       /* Used to control exclusive use of kd_buffer */
 
-#define DBG_FUNC_MASK 0xfffffffc
+#define DBG_FUNC_MASK  0xfffffffc
+
+/*  TODO: move to kdebug.h */
+#define CLASS_MASK      0xff000000
+#define CLASS_OFFSET    24
+#define SUBCLASS_MASK   0x00ff0000
+#define SUBCLASS_OFFSET 16
+#define CSC_MASK        0xffff0000     /*  class and subclass mask */
+#define CSC_OFFSET      SUBCLASS_OFFSET
+
+#define EXTRACT_CLASS(debugid)          ( (uint8_t) ( ((debugid) & CLASS_MASK   ) >> CLASS_OFFSET    ) )
+#define EXTRACT_SUBCLASS(debugid)       ( (uint8_t) ( ((debugid) & SUBCLASS_MASK) >> SUBCLASS_OFFSET ) )
+#define EXTRACT_CSC(debugid)            ( (uint16_t)( ((debugid) & CSC_MASK     ) >> CSC_OFFSET      ) )
+
+#define INTERRUPT      0x01050000
+#define MACH_vmfault   0x01300008
+#define BSC_SysCall    0x040c0000
+#define MACH_SysCall   0x010c0000
+#define DBG_SCALL_MASK 0xffff0000
 
 /* task to string structure */
 struct tts
@@ -166,174 +322,787 @@ typedef struct tts tts_t;
 
 struct krt
 {
-  kd_threadmap *map;    /* pointer to the map buffer */
-  int count;
-  int maxcount;
-  struct tts *atts;
+       kd_threadmap *map;    /* pointer to the map buffer */
+       int count;
+       int maxcount;
+       struct tts *atts;
 };
 
 typedef struct krt krt_t;
 
 /* This is for the CHUD toolkit call */
-typedef void (*kd_chudhook_fn) (unsigned int debugid, unsigned int arg1,
-                               unsigned int arg2, unsigned int arg3,
-                               unsigned int arg4, unsigned int arg5);
+typedef void (*kd_chudhook_fn) (uint32_t debugid, uintptr_t arg1,
+                               uintptr_t arg2, uintptr_t arg3,
+                               uintptr_t arg4, uintptr_t arg5);
 
-kd_chudhook_fn kdebug_chudhook = 0;   /* pointer to CHUD toolkit function */
+volatile kd_chudhook_fn kdebug_chudhook = 0;   /* pointer to CHUD toolkit function */
 
 __private_extern__ void stackshot_lock_init( void );
 
-/* Support syscall SYS_kdebug_trace */
-int
-kdebug_trace(__unused struct proc *p, struct kdebug_trace_args *uap, __unused register_t *retval)
+static uint8_t *type_filter_bitmap;
+
+/*
+ * This allows kperf to swap out the global state pid when kperf ownership is
+ * passed from one process to another. It checks the old global state pid so
+ * that kperf can't accidentally steal control of trace when a non-kperf trace user has
+ * control of trace.
+ */
+void
+kdbg_swap_global_state_pid(pid_t old_pid, pid_t new_pid);
+
+void
+kdbg_swap_global_state_pid(pid_t old_pid, pid_t new_pid)
 {
-    if ( (kdebug_enable == 0) )
-        return(EINVAL);
-  
-    kernel_debug(uap->code, uap->arg1, uap->arg2, uap->arg3, uap->arg4, 0);
-    return(0);
+       if (!(kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT))
+               return;
+
+       lck_mtx_lock(kd_trace_mtx_sysctl);
+
+       if (old_pid == global_state_pid)
+               global_state_pid = new_pid;
+
+       lck_mtx_unlock(kd_trace_mtx_sysctl);
 }
 
+static uint32_t
+kdbg_cpu_count(boolean_t early_trace)
+{
+       if (early_trace) {
+               /*
+                * we've started tracing before the IOKit has even
+                * started running... just use the static max value
+                */
+               return max_ncpus;
+       }
+
+       host_basic_info_data_t hinfo;
+       mach_msg_type_number_t count = HOST_BASIC_INFO_COUNT;
+       host_info((host_t)1 /* BSD_HOST */, HOST_BASIC_INFO, (host_info_t)&hinfo, &count);
+       assert(hinfo.logical_cpu_max > 0);
+       return hinfo.logical_cpu_max;
+}
 
+#if MACH_ASSERT
+#endif /* MACH_ASSERT */
+
+static void
+kdbg_iop_list_callback(kd_iop_t* iop, kd_callback_type type, void* arg)
+{
+       while (iop) {
+               iop->callback.func(iop->callback.context, type, arg);
+               iop = iop->next;
+       }
+}
+
+static void
+kdbg_set_tracing_enabled(boolean_t enabled, uint32_t trace_type)
+{
+       int s = ml_set_interrupts_enabled(FALSE);
+       lck_spin_lock(kds_spin_lock);
+       if (enabled) {
+               kdebug_enable |= trace_type;
+               kd_ctrl_page.kdebug_slowcheck &= ~SLOW_NOLOG;
+               kd_ctrl_page.enabled = 1;
+               commpage_update_kdebug_enable();
+       } else {
+               kdebug_enable &= ~(KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT);
+               kd_ctrl_page.kdebug_slowcheck |= SLOW_NOLOG;
+               kd_ctrl_page.enabled = 0;
+               commpage_update_kdebug_enable();
+       }
+       lck_spin_unlock(kds_spin_lock);
+       ml_set_interrupts_enabled(s);
+
+       if (enabled) {
+               kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_KDEBUG_ENABLED, NULL);
+       } else {
+               /*
+                * If you do not flush the IOP trace buffers, they can linger
+                * for a considerable period; consider code which disables and
+                * deallocates without a final sync flush.
+                */
+               kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_KDEBUG_DISABLED, NULL);
+               kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_SYNC_FLUSH, NULL);
+       }
+}
+
+static void
+kdbg_set_flags(int slowflag, int enableflag, boolean_t enabled)
+{
+       int s = ml_set_interrupts_enabled(FALSE);
+       lck_spin_lock(kds_spin_lock);
+
+       if (enabled) {
+               kd_ctrl_page.kdebug_slowcheck |= slowflag;
+               kdebug_enable |= enableflag;
+       } else {
+               kd_ctrl_page.kdebug_slowcheck &= ~slowflag;
+               kdebug_enable &= ~enableflag;
+       }
+        
+       lck_spin_unlock(kds_spin_lock);
+       ml_set_interrupts_enabled(s);
+}
+
+void
+disable_wrap(uint32_t *old_slowcheck, uint32_t *old_flags)
+{
+       int s = ml_set_interrupts_enabled(FALSE);
+       lck_spin_lock(kds_spin_lock);
+
+       *old_slowcheck = kd_ctrl_page.kdebug_slowcheck;
+       *old_flags = kd_ctrl_page.kdebug_flags;
+
+       kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED;
+       kd_ctrl_page.kdebug_flags |= KDBG_NOWRAP;
+
+       lck_spin_unlock(kds_spin_lock);
+       ml_set_interrupts_enabled(s);
+}
+
+void
+enable_wrap(uint32_t old_slowcheck, boolean_t lostevents)
+{
+       int s = ml_set_interrupts_enabled(FALSE);
+       lck_spin_lock(kds_spin_lock);
+
+       kd_ctrl_page.kdebug_flags &= ~KDBG_NOWRAP;
+
+       if ( !(old_slowcheck & SLOW_NOLOG))
+               kd_ctrl_page.kdebug_slowcheck &= ~SLOW_NOLOG;
+
+       if (lostevents == TRUE)
+               kd_ctrl_page.kdebug_flags |= KDBG_WRAPPED;
+
+       lck_spin_unlock(kds_spin_lock);
+       ml_set_interrupts_enabled(s);
+}
 
 static int
-create_buffers(void)
+create_buffers(boolean_t early_trace)
 {
-        unsigned int   cpu, i;
-       int     nentries;
+        int    i;
+       int     p_buffer_size;
+       int     f_buffer_size;
+       int     f_buffers;
+       int     error = 0;
+
+       /*
+        * For the duration of this allocation, trace code will only reference
+        * kdebug_iops. Any iops registered after this enabling will not be
+        * messaged until the buffers are reallocated.
+        *
+        * TLDR; Must read kd_iops once and only once!
+        */
+       kd_ctrl_page.kdebug_iops = kd_iops;
 
-       nentries = nkdbufs / kd_cpus;
-       kd_bufsize = nentries * sizeof(kd_buf);
 
-       bzero((char *)kdbip, sizeof(struct kd_bufinfo) * kd_cpus);
+       /*
+        * If the list is valid, it is sorted, newest -> oldest. Each iop entry
+        * has a cpu_id of "the older entry + 1", so the highest cpu_id will
+        * be the list head + 1.
+        */
+
+       kd_ctrl_page.kdebug_cpus = kd_ctrl_page.kdebug_iops ? kd_ctrl_page.kdebug_iops->cpu_id + 1 : kdbg_cpu_count(early_trace);
+
+       if (kmem_alloc(kernel_map, (vm_offset_t *)&kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus) != KERN_SUCCESS) {
+               error = ENOSPC;
+               goto out;
+       }
+
+       if (nkdbufs < (kd_ctrl_page.kdebug_cpus * EVENTS_PER_STORAGE_UNIT * MIN_STORAGE_UNITS_PER_CPU))
+               n_storage_units = kd_ctrl_page.kdebug_cpus * MIN_STORAGE_UNITS_PER_CPU;
+       else
+               n_storage_units = nkdbufs / EVENTS_PER_STORAGE_UNIT;
+
+       nkdbufs = n_storage_units * EVENTS_PER_STORAGE_UNIT;
+
+       f_buffers = n_storage_units / N_STORAGE_UNITS_PER_BUFFER;
+       n_storage_buffers = f_buffers;
+
+       f_buffer_size = N_STORAGE_UNITS_PER_BUFFER * sizeof(struct kd_storage);
+       p_buffer_size = (n_storage_units % N_STORAGE_UNITS_PER_BUFFER) * sizeof(struct kd_storage);
+
+       if (p_buffer_size)
+               n_storage_buffers++;
+
+       kd_bufs = NULL;
 
        if (kdcopybuf == 0) {
-               if (kmem_alloc(kernel_map, (unsigned int *)&kdcopybuf, (vm_size_t)KDCOPYBUF_SIZE) != KERN_SUCCESS)
-                       return ENOMEM;
+               if (kmem_alloc(kernel_map, (vm_offset_t *)&kdcopybuf, (vm_size_t)KDCOPYBUF_SIZE) != KERN_SUCCESS) {
+                       error = ENOSPC;
+                       goto out;
+               }
        }
-       for (cpu = 0; cpu < kd_cpus; cpu++) {
-               if (kmem_alloc(kernel_map, (unsigned int *)&kdbip[cpu].kd_buffer, kd_bufsize) != KERN_SUCCESS)
-                       break;
+       if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs, (vm_size_t)(n_storage_buffers * sizeof(struct kd_storage_buffers))) != KERN_SUCCESS) {
+               error = ENOSPC;
+               goto out;
        }
-       if (cpu < kd_cpus) {
-               for (i = 0; i < cpu; i++)
-                       kmem_free(kernel_map, (vm_offset_t)kdbip[i].kd_buffer, kd_bufsize);
-               kd_bufsize = 0;
+       bzero(kd_bufs, n_storage_buffers * sizeof(struct kd_storage_buffers));
 
-               kmem_free(kernel_map, (vm_offset_t)kdcopybuf, KDCOPYBUF_SIZE);
-               kdcopybuf = NULL;
-               
-               return(ENOMEM);
+       for (i = 0; i < f_buffers; i++) {
+               if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)f_buffer_size) != KERN_SUCCESS) {
+                       error = ENOSPC;
+                       goto out;
+               }
+               bzero(kd_bufs[i].kdsb_addr, f_buffer_size);
+
+               kd_bufs[i].kdsb_size = f_buffer_size;
        }
-       for (cpu = 0; cpu < kd_cpus; cpu++) {
-               kdbip[cpu].kd_bufptr = kdbip[cpu].kd_buffer;
-               kdbip[cpu].kd_buflast = &kdbip[cpu].kd_bufptr[nentries];
-               kdbip[cpu].kd_readlast = kdbip[cpu].kd_bufptr;
+       if (p_buffer_size) {
+               if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)p_buffer_size) != KERN_SUCCESS) {
+                       error = ENOSPC;
+                       goto out;
+               }
+               bzero(kd_bufs[i].kdsb_addr, p_buffer_size);
+
+               kd_bufs[i].kdsb_size = p_buffer_size;
        }
-       kdebug_flags |= KDBG_BUFINIT;
+       n_storage_units = 0;
 
-       return(0);
-}
+       for (i = 0; i < n_storage_buffers; i++) {
+               struct kd_storage *kds;
+               int     n_elements;
+               int     n;
+
+               n_elements = kd_bufs[i].kdsb_size / sizeof(struct kd_storage);
+               kds = kd_bufs[i].kdsb_addr;
 
+               for (n = 0; n < n_elements; n++) {
+                       kds[n].kds_next.buffer_index = kd_ctrl_page.kds_free_list.buffer_index;
+                       kds[n].kds_next.offset = kd_ctrl_page.kds_free_list.offset;
+
+                       kd_ctrl_page.kds_free_list.buffer_index = i;
+                       kd_ctrl_page.kds_free_list.offset = n;
+               }
+               n_storage_units += n_elements;
+       }
+
+       bzero((char *)kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus);
+
+       for (i = 0; i < (int)kd_ctrl_page.kdebug_cpus; i++) {
+               kdbip[i].kd_list_head.raw = KDS_PTR_NULL;
+               kdbip[i].kd_list_tail.raw = KDS_PTR_NULL;
+               kdbip[i].kd_lostevents = FALSE;
+               kdbip[i].num_bufs = 0;
+       }
+        
+       kd_ctrl_page.kdebug_flags |= KDBG_BUFINIT;
+
+       kd_ctrl_page.kds_inuse_count = 0;
+       n_storage_threshold = n_storage_units / 2;
+out:
+       if (error)
+               delete_buffers();
+
+       return(error);
+}
 
 static void
 delete_buffers(void)
 {
-        unsigned int   cpu;
+       int i;
+       
+       if (kd_bufs) {
+               for (i = 0; i < n_storage_buffers; i++) {
+                       if (kd_bufs[i].kdsb_addr) {
+                               kmem_free(kernel_map, (vm_offset_t)kd_bufs[i].kdsb_addr, (vm_size_t)kd_bufs[i].kdsb_size);
+                       }
+               }
+               kmem_free(kernel_map, (vm_offset_t)kd_bufs, (vm_size_t)(n_storage_buffers * sizeof(struct kd_storage_buffers)));
 
-       if (kd_bufsize && (kdebug_flags & KDBG_BUFINIT)) {
-               for (cpu = 0; cpu < kd_cpus; cpu++)
-                       kmem_free(kernel_map, (vm_offset_t)kdbip[cpu].kd_buffer, kd_bufsize);
-               kd_bufsize = 0;
+               kd_bufs = NULL;
+               n_storage_buffers = 0;
        }
        if (kdcopybuf) {
                kmem_free(kernel_map, (vm_offset_t)kdcopybuf, KDCOPYBUF_SIZE);
+
                kdcopybuf = NULL;
        }
-       kdebug_flags &= ~KDBG_BUFINIT;
+       kd_ctrl_page.kds_free_list.raw = KDS_PTR_NULL;
+
+       if (kdbip) {
+               kmem_free(kernel_map, (vm_offset_t)kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus);
+               
+               kdbip = NULL;
+       }
+        kd_ctrl_page.kdebug_iops = NULL;
+       kd_ctrl_page.kdebug_cpus = 0;
+       kd_ctrl_page.kdebug_flags &= ~KDBG_BUFINIT;
 }
 
+void
+release_storage_unit(int cpu, uint32_t kdsp_raw)
+{
+       int s = 0;
+       struct  kd_storage *kdsp_actual;
+       struct kd_bufinfo *kdbp;
+       union kds_ptr kdsp;
+
+       kdsp.raw = kdsp_raw;
 
-static void
-kernel_debug_internal(unsigned int debugid, unsigned int arg1, unsigned int arg2, unsigned int arg3,
-                     unsigned int arg4, unsigned int arg5, int entropy_flag)
+       s = ml_set_interrupts_enabled(FALSE);
+       lck_spin_lock(kds_spin_lock);
+
+       kdbp = &kdbip[cpu];
+
+       if (kdsp.raw == kdbp->kd_list_head.raw) {
+               /*
+                * it's possible for the storage unit pointed to
+                * by kdsp to have already been stolen... so
+                * check to see if it's still the head of the list
+                * now that we're behind the lock that protects 
+                * adding and removing from the queue...
+                * since we only ever release and steal units from
+                * that position, if it's no longer the head
+                * we having nothing to do in this context
+                */
+               kdsp_actual = POINTER_FROM_KDS_PTR(kdsp);
+               kdbp->kd_list_head = kdsp_actual->kds_next;
+
+               kdsp_actual->kds_next = kd_ctrl_page.kds_free_list;
+               kd_ctrl_page.kds_free_list = kdsp;
+
+               kd_ctrl_page.kds_inuse_count--;
+       }
+       lck_spin_unlock(kds_spin_lock);
+       ml_set_interrupts_enabled(s);
+}
+
+
+boolean_t
+allocate_storage_unit(int cpu)
 {
-       int s;
-       kd_buf * kd;
-       struct proc *curproc;
-       unsigned long long now;
-       int cpu;
-       
+       union   kds_ptr kdsp;
+       struct  kd_storage *kdsp_actual, *kdsp_next_actual;
+       struct  kd_bufinfo *kdbp, *kdbp_vict, *kdbp_try;
+       uint64_t        oldest_ts, ts;
+       boolean_t       retval = TRUE;
+       int                     s = 0;
+               
        s = ml_set_interrupts_enabled(FALSE);
+       lck_spin_lock(kds_spin_lock);
 
-       now = mach_absolute_time() & KDBG_TIMESTAMP_MASK;
-       cpu = cpu_number();
+       kdbp = &kdbip[cpu];
 
-       if (kdebug_enable & KDEBUG_ENABLE_CHUD) {
-           if (kdebug_chudhook)
-               kdebug_chudhook(debugid, arg1, arg2, arg3, arg4, arg5);
+       /* If someone beat us to the allocate, return success */
+       if (kdbp->kd_list_tail.raw != KDS_PTR_NULL) {
+               kdsp_actual = POINTER_FROM_KDS_PTR(kdbp->kd_list_tail);
 
-           if ( !(kdebug_enable & (KDEBUG_ENABLE_ENTROPY | KDEBUG_ENABLE_TRACE)))
-               goto out;
+               if (kdsp_actual->kds_bufindx < EVENTS_PER_STORAGE_UNIT)
+                       goto out;
        }
+       
+       if ((kdsp = kd_ctrl_page.kds_free_list).raw != KDS_PTR_NULL) {
+               kdsp_actual = POINTER_FROM_KDS_PTR(kdsp);
+               kd_ctrl_page.kds_free_list = kdsp_actual->kds_next;
+
+               kd_ctrl_page.kds_inuse_count++;
+       } else {
+               if (kd_ctrl_page.kdebug_flags & KDBG_NOWRAP) {
+                       kd_ctrl_page.kdebug_slowcheck |= SLOW_NOLOG;
+                       kdbp->kd_lostevents = TRUE;
+                       retval = FALSE;
+                       goto out;
+               }
+               kdbp_vict = NULL;
+               oldest_ts = (uint64_t)-1;
 
-       if (kdebug_slowcheck == 0)
-           goto record_trace;
+               for (kdbp_try = &kdbip[0]; kdbp_try < &kdbip[kd_ctrl_page.kdebug_cpus]; kdbp_try++) {
 
-       if (entropy_flag && (kdebug_enable & KDEBUG_ENABLE_ENTROPY))
-         {
-           if (kd_entropy_indx < kd_entropy_count)
-             {
-               kd_entropy_buffer [ kd_entropy_indx] = mach_absolute_time();
-               kd_entropy_indx++;
-             }
-           
-           if (kd_entropy_indx == kd_entropy_count)
-             {
-               /* Disable entropy collection */
-               kdebug_enable &= ~KDEBUG_ENABLE_ENTROPY;
-               kdebug_slowcheck &= ~SLOW_ENTROPY;
-             }
-         }
-
-       if ( (kdebug_slowcheck & SLOW_NOLOG) )
-           goto out;
+                       if (kdbp_try->kd_list_head.raw == KDS_PTR_NULL) {
+                               /*
+                                * no storage unit to steal
+                                */
+                               continue;
+                       }
+
+                       kdsp_actual = POINTER_FROM_KDS_PTR(kdbp_try->kd_list_head);
+
+                       if (kdsp_actual->kds_bufcnt < EVENTS_PER_STORAGE_UNIT) {
+                               /*
+                                * make sure we don't steal the storage unit
+                                * being actively recorded to...  need to
+                                * move on because we don't want an out-of-order
+                                * set of events showing up later
+                                */
+                               continue;
+                       }
+                       ts = kdbg_get_timestamp(&kdsp_actual->kds_records[0]);
+
+                       if (ts < oldest_ts) {
+                               /*
+                                * when 'wrapping', we want to steal the
+                                * storage unit that has the 'earliest' time
+                                * associated with it (first event time)
+                                */
+                               oldest_ts = ts;
+                               kdbp_vict = kdbp_try;
+                       }
+               }
+               if (kdbp_vict == NULL) {
+                       kdebug_enable = 0;
+                       kd_ctrl_page.enabled = 0;
+                       commpage_update_kdebug_enable();
+                       retval = FALSE;
+                       goto out;
+               }
+               kdsp = kdbp_vict->kd_list_head;
+               kdsp_actual = POINTER_FROM_KDS_PTR(kdsp);
+               kdbp_vict->kd_list_head = kdsp_actual->kds_next;
+
+               if (kdbp_vict->kd_list_head.raw != KDS_PTR_NULL) {
+                       kdsp_next_actual = POINTER_FROM_KDS_PTR(kdbp_vict->kd_list_head);
+                       kdsp_next_actual->kds_lostevents = TRUE;
+               } else
+                       kdbp_vict->kd_lostevents = TRUE;
+
+               kd_ctrl_page.kdebug_flags |= KDBG_WRAPPED;
+       }
+       kdsp_actual->kds_timestamp = mach_absolute_time();
+       kdsp_actual->kds_next.raw = KDS_PTR_NULL;
+       kdsp_actual->kds_bufcnt   = 0;
+       kdsp_actual->kds_readlast = 0;
+
+       kdsp_actual->kds_lostevents = kdbp->kd_lostevents;
+       kdbp->kd_lostevents = FALSE;
+       kdsp_actual->kds_bufindx  = 0;
+
+       if (kdbp->kd_list_head.raw == KDS_PTR_NULL)
+               kdbp->kd_list_head = kdsp;
+       else
+               POINTER_FROM_KDS_PTR(kdbp->kd_list_tail)->kds_next = kdsp;
+       kdbp->kd_list_tail = kdsp;
+out:
+       lck_spin_unlock(kds_spin_lock);
+       ml_set_interrupts_enabled(s);
+
+       return (retval);
+}
+
+int
+kernel_debug_register_callback(kd_callback_t callback)
+{
+       kd_iop_t* iop;
+       if (kmem_alloc(kernel_map, (vm_offset_t *)&iop, sizeof(kd_iop_t)) == KERN_SUCCESS) {
+               memcpy(&iop->callback, &callback, sizeof(kd_callback_t));
+               
+               /*
+                * <rdar://problem/13351477> Some IOP clients are not providing a name.
+                *
+                * Remove when fixed.
+                */
+               {
+                       boolean_t is_valid_name = FALSE;
+                       for (uint32_t length=0; length<sizeof(callback.iop_name); ++length) {
+                               /* This is roughly isprintable(c) */
+                               if (callback.iop_name[length] > 0x20 && callback.iop_name[length] < 0x7F)
+                                       continue;
+                               if (callback.iop_name[length] == 0) {
+                                       if (length)
+                                               is_valid_name = TRUE;
+                                       break;
+                               }
+                       }
+                       
+                       if (!is_valid_name) {
+                               strlcpy(iop->callback.iop_name, "IOP-???", sizeof(iop->callback.iop_name));
+                       }
+               }
+
+               iop->last_timestamp = 0;
+               
+               do {
+                       /*
+                        * We use two pieces of state, the old list head
+                        * pointer, and the value of old_list_head->cpu_id.
+                        * If we read kd_iops more than once, it can change
+                        * between reads.
+                        *
+                        * TLDR; Must not read kd_iops more than once per loop.
+                        */
+                       iop->next = kd_iops;
+                       iop->cpu_id = iop->next ? (iop->next->cpu_id+1) : kdbg_cpu_count(FALSE);
+
+                       /*
+                        * Header says OSCompareAndSwapPtr has a memory barrier
+                        */
+               } while (!OSCompareAndSwapPtr(iop->next, iop, (void* volatile*)&kd_iops));
+
+               return iop->cpu_id;
+       }
+
+       return 0;
+}
+
+void
+kernel_debug_enter(
+       uint32_t        coreid,
+       uint32_t        debugid,
+       uint64_t        timestamp,
+       uintptr_t       arg1,
+       uintptr_t       arg2,
+       uintptr_t       arg3,
+       uintptr_t       arg4,
+       uintptr_t       threadid
+       )
+{
+       uint32_t        bindx;
+       kd_buf          *kd;
+       struct kd_bufinfo *kdbp;
+       struct kd_storage *kdsp_actual;
+       union  kds_ptr kds_raw;
+
+       if (kd_ctrl_page.kdebug_slowcheck) {
+
+               if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) || !(kdebug_enable & (KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT)))
+                       goto out1;
        
-       if (kdebug_flags & KDBG_PIDCHECK)
-         {
-           /* If kdebug flag is not set for current proc, return  */
-           curproc = current_proc();
-           if ((curproc && !(curproc->p_flag & P_KDEBUG)) &&
-               ((debugid&0xffff0000) != (MACHDBG_CODE(DBG_MACH_SCHED, 0) | DBG_FUNC_NONE)))
-               goto out;
-         }
-       else if (kdebug_flags & KDBG_PIDEXCLUDE)
-         {
-           /* If kdebug flag is set for current proc, return  */
-           curproc = current_proc();
-           if ((curproc && (curproc->p_flag & P_KDEBUG)) &&
-               ((debugid&0xffff0000) != (MACHDBG_CODE(DBG_MACH_SCHED, 0) | DBG_FUNC_NONE)))
-               goto out;
-         }
+               if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) {
+                       if (isset(type_filter_bitmap, EXTRACT_CSC(debugid))) 
+                               goto record_event;
+                       goto out1;
+               }
+               else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) {
+                       if (debugid >= kdlog_beg && debugid <= kdlog_end)
+                               goto record_event;
+                       goto out1;
+               }
+               else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) {
+                       if ((debugid & DBG_FUNC_MASK) != kdlog_value1 &&
+                               (debugid & DBG_FUNC_MASK) != kdlog_value2 &&
+                               (debugid & DBG_FUNC_MASK) != kdlog_value3 &&
+                               (debugid & DBG_FUNC_MASK) != kdlog_value4)
+                               goto out1;
+               }
+       }
+       
+record_event:
+
+       disable_preemption();
 
-       if (kdebug_flags & KDBG_RANGECHECK)
-         {
-           if ((debugid < kdlog_beg)
-               || ((debugid >= kdlog_end) && (debugid >> 24 != DBG_TRACE)))
+       if (kd_ctrl_page.enabled == 0)
                goto out;
-         }
-       else if (kdebug_flags & KDBG_VALCHECK)
-         {
-           if ((debugid & DBG_FUNC_MASK) != kdlog_value1 &&
-               (debugid & DBG_FUNC_MASK) != kdlog_value2 &&
-               (debugid & DBG_FUNC_MASK) != kdlog_value3 &&
-               (debugid & DBG_FUNC_MASK) != kdlog_value4 &&
-               (debugid >> 24 != DBG_TRACE))
+
+       kdbp = &kdbip[coreid];
+       timestamp &= KDBG_TIMESTAMP_MASK;
+
+#if KDEBUG_MOJO_TRACE
+       if (kdebug_enable & KDEBUG_ENABLE_SERIAL)
+               kdebug_serial_print(coreid, debugid, timestamp,
+                                   arg1, arg2, arg3, arg4, threadid);
+#endif
+
+retry_q:
+       kds_raw = kdbp->kd_list_tail;
+
+       if (kds_raw.raw != KDS_PTR_NULL) {
+               kdsp_actual = POINTER_FROM_KDS_PTR(kds_raw);
+               bindx = kdsp_actual->kds_bufindx;
+       } else
+               kdsp_actual = NULL;
+       
+       if (kdsp_actual == NULL || bindx >= EVENTS_PER_STORAGE_UNIT) {
+               if (allocate_storage_unit(coreid) == FALSE) {
+                       /*
+                        * this can only happen if wrapping
+                        * has been disabled
+                        */
+                       goto out;
+               }
+               goto retry_q;
+       }
+       if ( !OSCompareAndSwap(bindx, bindx + 1, &kdsp_actual->kds_bufindx))
+               goto retry_q;
+
+       // IOP entries can be allocated before xnu allocates and inits the buffer
+       if (timestamp < kdsp_actual->kds_timestamp)
+               kdsp_actual->kds_timestamp = timestamp;
+
+       kd = &kdsp_actual->kds_records[bindx];
+
+       kd->debugid = debugid;
+       kd->arg1 = arg1;
+       kd->arg2 = arg2;
+       kd->arg3 = arg3;
+       kd->arg4 = arg4;
+       kd->arg5 = threadid;
+                 
+       kdbg_set_timestamp_and_cpu(kd, timestamp, coreid);
+
+       OSAddAtomic(1, &kdsp_actual->kds_bufcnt);
+out:
+       enable_preemption();
+out1:
+       if ((kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold)) {
+               boolean_t need_kds_wakeup = FALSE;
+               int     s;
+
+               /*
+                * try to take the lock here to synchronize with the
+                * waiter entering the blocked state... use the try
+                * mode to prevent deadlocks caused by re-entering this
+                * routine due to various trace points triggered in the
+                * lck_spin_sleep_xxxx routines used to actually enter
+                * our wait condition... no problem if we fail,
+                * there will be lots of additional events coming in that
+                * will eventually succeed in grabbing this lock
+                */
+               s = ml_set_interrupts_enabled(FALSE);
+
+               if (lck_spin_try_lock(kdw_spin_lock)) {
+
+                       if (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) {
+                               kds_waiter = 0;
+                               need_kds_wakeup = TRUE;
+                       }
+                       lck_spin_unlock(kdw_spin_lock);
+
+                       ml_set_interrupts_enabled(s);
+                       
+                       if (need_kds_wakeup == TRUE)
+                               wakeup(&kds_waiter);
+               }
+       }
+}
+
+
+
+static void
+kernel_debug_internal(
+       uint32_t        debugid,
+       uintptr_t       arg1,
+       uintptr_t       arg2,
+       uintptr_t       arg3,
+       uintptr_t       arg4,
+       uintptr_t       arg5)
+{
+       struct proc     *curproc;
+       uint64_t        now;
+       uint32_t        bindx;
+       boolean_t       s;
+       kd_buf          *kd;
+       int             cpu;
+       struct kd_bufinfo *kdbp;
+       struct kd_storage *kdsp_actual;
+       union  kds_ptr kds_raw;
+
+       
+
+       if (kd_ctrl_page.kdebug_slowcheck) {
+
+               if (kdebug_enable & KDEBUG_ENABLE_CHUD) {
+                       kd_chudhook_fn chudhook;
+                       /*
+                        * Mask interrupts to minimize the interval across
+                        * which the driver providing the hook could be
+                        * unloaded.
+                        */
+                       s = ml_set_interrupts_enabled(FALSE);
+                       chudhook = kdebug_chudhook;
+                       if (chudhook)
+                               chudhook(debugid, arg1, arg2, arg3, arg4, arg5);
+                       ml_set_interrupts_enabled(s);
+               }
+               if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) || !(kdebug_enable & (KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT)))
+                       goto out1;
+       
+               if ( !ml_at_interrupt_context()) {
+                       if (kd_ctrl_page.kdebug_flags & KDBG_PIDCHECK) {
+                               /*
+                                * If kdebug flag is not set for current proc, return
+                                */
+                               curproc = current_proc();
+
+                               if ((curproc && !(curproc->p_kdebug)) &&
+                                   ((debugid & 0xffff0000) != (MACHDBG_CODE(DBG_MACH_SCHED, 0) | DBG_FUNC_NONE)) &&
+                                     (debugid >> 24 != DBG_TRACE))
+                                       goto out1;
+                       }
+                       else if (kd_ctrl_page.kdebug_flags & KDBG_PIDEXCLUDE) {
+                               /*
+                                * If kdebug flag is set for current proc, return
+                                */
+                               curproc = current_proc();
+
+                               if ((curproc && curproc->p_kdebug) &&
+                                   ((debugid & 0xffff0000) != (MACHDBG_CODE(DBG_MACH_SCHED, 0) | DBG_FUNC_NONE)) &&
+                                     (debugid >> 24 != DBG_TRACE))
+                                       goto out1;
+                       }
+               }
+
+               if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) {
+                       /* Always record trace system info */
+                       if (EXTRACT_CLASS(debugid) == DBG_TRACE)
+                               goto record_event;
+
+                       if (isset(type_filter_bitmap, EXTRACT_CSC(debugid))) 
+                               goto record_event;
+                       goto out1;
+               }
+               else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) {
+                       /* Always record trace system info */
+                       if (EXTRACT_CLASS(debugid) == DBG_TRACE)
+                               goto record_event;
+                               
+                       if (debugid < kdlog_beg || debugid > kdlog_end)
+                               goto out1;
+               }
+               else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) {
+                       /* Always record trace system info */
+                       if (EXTRACT_CLASS(debugid) == DBG_TRACE)
+                               goto record_event;
+               
+                       if ((debugid & DBG_FUNC_MASK) != kdlog_value1 &&
+                           (debugid & DBG_FUNC_MASK) != kdlog_value2 &&
+                           (debugid & DBG_FUNC_MASK) != kdlog_value3 &&
+                           (debugid & DBG_FUNC_MASK) != kdlog_value4)
+                               goto out1;
+               }
+       }
+record_event:
+       disable_preemption();
+
+       if (kd_ctrl_page.enabled == 0)
                goto out;
-         }
 
-record_trace:
-       kd = kdbip[cpu].kd_bufptr;
+       cpu = cpu_number();
+       kdbp = &kdbip[cpu];
+
+#if KDEBUG_MOJO_TRACE
+       if (kdebug_enable & KDEBUG_ENABLE_SERIAL)
+               kdebug_serial_print(cpu, debugid,
+                                   mach_absolute_time() & KDBG_TIMESTAMP_MASK,
+                                   arg1, arg2, arg3, arg4, arg5);
+#endif
+
+retry_q:
+       kds_raw = kdbp->kd_list_tail;
+
+       if (kds_raw.raw != KDS_PTR_NULL) {
+               kdsp_actual = POINTER_FROM_KDS_PTR(kds_raw);
+               bindx = kdsp_actual->kds_bufindx;
+       } else
+               kdsp_actual = NULL;
+       
+       if (kdsp_actual == NULL || bindx >= EVENTS_PER_STORAGE_UNIT) {
+               if (allocate_storage_unit(cpu) == FALSE) {
+                       /*
+                        * this can only happen if wrapping
+                        * has been disabled
+                        */
+                       goto out;
+               }
+               goto retry_q;
+       }
+       now = mach_absolute_time() & KDBG_TIMESTAMP_MASK;
+
+       if ( !OSCompareAndSwap(bindx, bindx + 1, &kdsp_actual->kds_bufindx))
+               goto retry_q;
+
+       kd = &kdsp_actual->kds_records[bindx];
+
        kd->debugid = debugid;
        kd->arg1 = arg1;
        kd->arg2 = arg2;
@@ -341,70 +1110,216 @@ record_trace:
        kd->arg4 = arg4;
        kd->arg5 = arg5;
                  
-       /*
-        * Watch for out of order timestamps
-        */     
-       if (now < kdbip[cpu].kd_prev_timebase)
-         {
-           /*
-            * if so, just store the previous timestamp + a cycle
-            */
-           now = ++kdbip[cpu].kd_prev_timebase & KDBG_TIMESTAMP_MASK;
-         }
-       else
-         {
-           kdbip[cpu].kd_prev_timebase = now;
-         }
-       kd->timestamp = now | (((uint64_t)cpu) << KDBG_CPU_SHIFT);
+       kdbg_set_timestamp_and_cpu(kd, now, cpu);
+
+       OSAddAtomic(1, &kdsp_actual->kds_bufcnt);
+out:
+       enable_preemption();
+out1:
+       if (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) {
+               uint32_t        etype;
+               uint32_t        stype;
+               
+               etype = debugid & DBG_FUNC_MASK;
+               stype = debugid & DBG_SCALL_MASK;
+
+               if (etype == INTERRUPT || etype == MACH_vmfault ||
+                   stype == BSC_SysCall || stype == MACH_SysCall) {
+
+                       boolean_t need_kds_wakeup = FALSE;
+
+                       /*
+                        * try to take the lock here to synchronize with the
+                        * waiter entering the blocked state... use the try
+                        * mode to prevent deadlocks caused by re-entering this
+                        * routine due to various trace points triggered in the
+                        * lck_spin_sleep_xxxx routines used to actually enter
+                        * one of our 2 wait conditions... no problem if we fail,
+                        * there will be lots of additional events coming in that
+                        * will eventually succeed in grabbing this lock
+                        */
+                       s = ml_set_interrupts_enabled(FALSE);
+
+                       if (lck_spin_try_lock(kdw_spin_lock)) {
+
+                               if (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) {
+                                       kds_waiter = 0;
+                                       need_kds_wakeup = TRUE;
+                               }
+                               lck_spin_unlock(kdw_spin_lock);
+                       }
+                       ml_set_interrupts_enabled(s);
+                       
+                       if (need_kds_wakeup == TRUE)
+                               wakeup(&kds_waiter);
+               }
+       }
+}
+
+void
+kernel_debug(
+       uint32_t        debugid,
+       uintptr_t       arg1,
+       uintptr_t       arg2,
+       uintptr_t       arg3,
+       uintptr_t       arg4,
+       __unused uintptr_t arg5)
+{
+       kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, (uintptr_t)thread_tid(current_thread()));
+}
+
+void
+kernel_debug1(
+       uint32_t        debugid,
+       uintptr_t       arg1,
+       uintptr_t       arg2,
+       uintptr_t       arg3,
+       uintptr_t       arg4,
+       uintptr_t       arg5)
+{
+       kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, arg5);
+}
+
+void
+kernel_debug_string(const char *message)
+{
+       uintptr_t arg[4] = {0, 0, 0, 0};
+
+       /* Stuff the message string in the args and log it. */
+        strncpy((char *)arg, message, MIN(sizeof(arg), strlen(message)));
+       KERNEL_DEBUG_EARLY(
+               TRACE_INFO_STRING,
+               arg[0], arg[1], arg[2], arg[3]);
+}
+
+extern int     master_cpu;             /* MACH_KERNEL_PRIVATE */
+/*
+ * Used prior to start_kern_tracing() being called.
+ * Log temporarily into a static buffer.
+ */
+void
+kernel_debug_early(
+       uint32_t        debugid,
+       uintptr_t       arg1,
+       uintptr_t       arg2,
+       uintptr_t       arg3,
+       uintptr_t       arg4)
+{
+       /* If tracing is already initialized, use it */
+       if (nkdbufs) {
+               KERNEL_DEBUG_CONSTANT(debugid, arg1, arg2, arg3, arg4, 0);
+               return;
+       }
+
+       /* Do nothing if the buffer is full or we're not on the boot cpu */ 
+       kd_early_overflow = kd_early_index >= KD_EARLY_BUFFER_MAX;
+       if (kd_early_overflow ||
+           cpu_number() != master_cpu)
+               return;
+
+       kd_early_buffer[kd_early_index].debugid = debugid;
+       kd_early_buffer[kd_early_index].timestamp = mach_absolute_time();
+       kd_early_buffer[kd_early_index].arg1 = arg1;
+       kd_early_buffer[kd_early_index].arg2 = arg2;
+       kd_early_buffer[kd_early_index].arg3 = arg3;
+       kd_early_buffer[kd_early_index].arg4 = arg4;
+       kd_early_buffer[kd_early_index].arg5 = 0;
+       kd_early_index++;
+}
+
+/*
+ * Transfen the contents of the temporary buffer into the trace buffers.
+ * Precede that by logging the rebase time (offset) - the TSC-based time (in ns)
+ * when mach_absolute_time is set to 0.
+ */
+static void
+kernel_debug_early_end(void)
+{
+       int     i;
+
+       if (cpu_number() != master_cpu)
+               panic("kernel_debug_early_end() not call on boot processor");
+
+       /* Fake sentinel marking the start of kernel time relative to TSC */
+       kernel_debug_enter(
+               0,
+               TRACE_TIMESTAMPS,
+               0,
+               (uint32_t)(tsc_rebase_abs_time >> 32),
+               (uint32_t)tsc_rebase_abs_time,
+               0,
+               0,
+               0);
+       for (i = 0; i < kd_early_index; i++) {
+               kernel_debug_enter(
+                       0,
+                       kd_early_buffer[i].debugid,
+                       kd_early_buffer[i].timestamp,
+                       kd_early_buffer[i].arg1,
+                       kd_early_buffer[i].arg2,
+                       kd_early_buffer[i].arg3,
+                       kd_early_buffer[i].arg4,
+                       0);
+       }
+
+       /* Cut events-lost event on overflow */
+       if (kd_early_overflow)
+               KERNEL_DEBUG_CONSTANT(
+                       TRACE_LOST_EVENTS, 0, 0, 0, 0, 0);
+
+       /* This trace marks the start of kernel tracing */
+       kernel_debug_string("early trace done");
+}
+
+/*
+ * Support syscall SYS_kdebug_trace. U64->K32 args may get truncated in kdebug_trace64
+ */
+int
+kdebug_trace(struct proc *p, struct kdebug_trace_args *uap, int32_t *retval)
+{
+       struct kdebug_trace64_args uap64;
+
+       uap64.code = uap->code;
+       uap64.arg1 = uap->arg1;
+       uap64.arg2 = uap->arg2;
+       uap64.arg3 = uap->arg3;
+       uap64.arg4 = uap->arg4;
+
+       return kdebug_trace64(p, &uap64, retval);
+}
 
-       kdbip[cpu].kd_bufptr++;
+/*
+ * Support syscall SYS_kdebug_trace64. 64-bit args on K32 will get truncated to fit in 32-bit record format.
+ */
+int kdebug_trace64(__unused struct proc *p, struct kdebug_trace64_args *uap, __unused int32_t *retval)
+{
+       uint8_t code_class;
 
-       if (kdbip[cpu].kd_bufptr >= kdbip[cpu].kd_buflast)
-               kdbip[cpu].kd_bufptr = kdbip[cpu].kd_buffer;
+       /*
+        * Not all class are supported for injection from userspace, especially ones used by the core
+        * kernel tracing infrastructure.
+        */
+       code_class = EXTRACT_CLASS(uap->code);
 
-       if (kdbip[cpu].kd_bufptr == kdbip[cpu].kd_readlast) {
-               if (kdebug_flags & KDBG_NOWRAP)
-                       kdebug_slowcheck |= SLOW_NOLOG;
-               kdbip[cpu].kd_wrapped = 1;
-               kdebug_flags |= KDBG_WRAPPED;
+       switch (code_class) {
+               case DBG_TRACE:
+                       return EPERM;
        }
 
-out:
-       ml_set_interrupts_enabled(s);
-}
+       if ( __probable(kdebug_enable == 0) )
+               return(0); 
 
-void
-kernel_debug(unsigned int debugid, unsigned int arg1, unsigned int arg2, unsigned int arg3,
-                     unsigned int arg4, __unused unsigned int arg5)
-{
-       kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, (int)current_thread(), 1);
-}
+       kernel_debug_internal(uap->code, (uintptr_t)uap->arg1, (uintptr_t)uap->arg2, (uintptr_t)uap->arg3, (uintptr_t)uap->arg4, (uintptr_t)thread_tid(current_thread()));
 
-void
-kernel_debug1(unsigned int debugid, unsigned int arg1, unsigned int arg2, unsigned int arg3,
-                     unsigned int arg4, unsigned int arg5)
-{
-       kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, arg5, 0);
+       return(0);
 }
 
 static void
 kdbg_lock_init(void)
 {
-       host_basic_info_data_t hinfo;
-       mach_msg_type_number_t count = HOST_BASIC_INFO_COUNT;
-
-
-       if (kdebug_flags & KDBG_LOCKINIT)
+       if (kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT)
                return;
-
-       /* get the number of cpus and cache it */
-#define BSD_HOST 1
-       host_info((host_t)BSD_HOST, HOST_BASIC_INFO, (host_info_t)&hinfo, &count);
-       kd_cpus = hinfo.physical_cpu_max;
-
-       if (kmem_alloc(kernel_map, (unsigned int *)&kdbip, sizeof(struct kd_bufinfo) * kd_cpus) != KERN_SUCCESS)
-               return;
-
+       
        /*
         * allocate lock group attribute and group
         */
@@ -418,232 +1333,318 @@ kdbg_lock_init(void)
 
 
        /*
-        * allocate and initialize spin lock and mutex
+        * allocate and initialize mutex's
         */
        kd_trace_mtx_sysctl = lck_mtx_alloc_init(kd_trace_mtx_sysctl_grp, kd_trace_mtx_sysctl_attr);
+       kds_spin_lock = lck_spin_alloc_init(kd_trace_mtx_sysctl_grp, kd_trace_mtx_sysctl_attr);
+       kdw_spin_lock = lck_spin_alloc_init(kd_trace_mtx_sysctl_grp, kd_trace_mtx_sysctl_attr);
 
-       kdebug_flags |= KDBG_LOCKINIT;
+       kd_ctrl_page.kdebug_flags |= KDBG_LOCKINIT;
 }
 
 
 int
-kdbg_bootstrap(void)
+kdbg_bootstrap(boolean_t early_trace)
 {
-        kdebug_flags &= ~KDBG_WRAPPED;
+        kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED;
 
-       return (create_buffers());
+       return (create_buffers(early_trace));
 }
 
 int
-kdbg_reinit(void)
+kdbg_reinit(boolean_t early_trace)
 {
-    int ret=0;
-
-    /*
-     * Disable trace collecting
-     * First make sure we're not in
-     * the middle of cutting a trace
-     */
-
-    kdebug_enable &= ~KDEBUG_ENABLE_TRACE;
-    kdebug_slowcheck |= SLOW_NOLOG;
-
-    /*
-     * make sure the SLOW_NOLOG is seen
-     * by everyone that might be trying
-     * to cut a trace..
-     */
-    IOSleep(100);
-
-    delete_buffers();
-
-    if ((kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr)
-      {
-       kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize);
-       kdebug_flags &= ~KDBG_MAPINIT;
-       kd_mapsize = 0;
-       kd_mapptr = (kd_threadmap *) 0;
-       kd_mapcount = 0;
-      }  
+       int ret = 0;
+
+       /*
+        * Disable trace collecting
+        * First make sure we're not in
+        * the middle of cutting a trace
+        */
+       kdbg_set_tracing_enabled(FALSE, KDEBUG_ENABLE_TRACE);
+
+       /*
+        * make sure the SLOW_NOLOG is seen
+        * by everyone that might be trying
+        * to cut a trace..
+        */
+       IOSleep(100);
+
+       delete_buffers();
+
+       if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr) {
+               kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize);
+               kd_ctrl_page.kdebug_flags &= ~KDBG_MAPINIT;
+               kd_mapsize = 0;
+               kd_mapptr = (kd_threadmap *) 0;
+               kd_mapcount = 0;
+       }  
+       ret = kdbg_bootstrap(early_trace);
 
-    ret = kdbg_bootstrap();
+       RAW_file_offset = 0;
+       RAW_file_written = 0;
 
-    return(ret);
+       return(ret);
 }
 
 void
 kdbg_trace_data(struct proc *proc, long *arg_pid)
 {
-    if (!proc)
-        *arg_pid = 0;
-    else
-       *arg_pid = proc->p_pid;
-    
-    return;
+       if (!proc)
+               *arg_pid = 0;
+       else
+               *arg_pid = proc->p_pid;
 }
 
 
 void
 kdbg_trace_string(struct proc *proc, long *arg1, long *arg2, long *arg3, long *arg4)
 {
-    int i;
-    char *dbg_nameptr; 
-    int dbg_namelen;
-    long dbg_parms[4];
-
-    if (!proc)
-      {
-       *arg1 = 0;
-       *arg2 = 0;
-       *arg3 = 0;
-       *arg4 = 0;
-       return;
-      }
-
-    /* Collect the pathname for tracing */
-    dbg_nameptr = proc->p_comm;
-    dbg_namelen = strlen(proc->p_comm);
-    dbg_parms[0]=0L;
-    dbg_parms[1]=0L;
-    dbg_parms[2]=0L;
-    dbg_parms[3]=0L;
+       char *dbg_nameptr; 
+       int dbg_namelen;
+       long dbg_parms[4];
+
+       if (!proc) {
+               *arg1 = 0;
+               *arg2 = 0;
+               *arg3 = 0;
+               *arg4 = 0;
+               return;
+       }
+       /*
+        * Collect the pathname for tracing
+        */
+       dbg_nameptr = proc->p_comm;
+       dbg_namelen = (int)strlen(proc->p_comm);
+       dbg_parms[0]=0L;
+       dbg_parms[1]=0L;
+       dbg_parms[2]=0L;
+       dbg_parms[3]=0L;
   
-    if(dbg_namelen > (int)sizeof(dbg_parms))
-      dbg_namelen = sizeof(dbg_parms);
+       if(dbg_namelen > (int)sizeof(dbg_parms))
+               dbg_namelen = (int)sizeof(dbg_parms);
     
-    for(i=0;dbg_namelen > 0; i++)
-      {
-       dbg_parms[i]=*(long*)dbg_nameptr;
-       dbg_nameptr += sizeof(long);
-       dbg_namelen -= sizeof(long);
-      }
-
-    *arg1=dbg_parms[0];
-    *arg2=dbg_parms[1];
-    *arg3=dbg_parms[2];
-    *arg4=dbg_parms[3];
+       strncpy((char *)dbg_parms, dbg_nameptr, dbg_namelen);
+
+       *arg1=dbg_parms[0];
+       *arg2=dbg_parms[1];
+       *arg3=dbg_parms[2];
+       *arg4=dbg_parms[3];
 }
 
 static void
 kdbg_resolve_map(thread_t th_act, void *opaque)
 {
-  kd_threadmap *mapptr;
-  krt_t *t = (krt_t *)opaque;
-
-  if(t->count < t->maxcount)
-    {
-      mapptr=&t->map[t->count];
-      mapptr->thread  = (unsigned int)th_act;
-      (void) strncpy (mapptr->command, t->atts->task_comm,
-                     sizeof(t->atts->task_comm)-1);
-      mapptr->command[sizeof(t->atts->task_comm)-1] = '\0';
-
-      /*
-       Some kernel threads have no associated pid.
-       We still need to mark the entry as valid.
-      */
-      if (t->atts->pid)
-         mapptr->valid = t->atts->pid;
-      else
-         mapptr->valid = 1;
-
-      t->count++;
-    }
+       kd_threadmap *mapptr;
+       krt_t *t = (krt_t *)opaque;
+
+       if (t->count < t->maxcount) {
+               mapptr = &t->map[t->count];
+               mapptr->thread  = (uintptr_t)thread_tid(th_act);
+
+               (void) strlcpy (mapptr->command, t->atts->task_comm,
+                               sizeof(t->atts->task_comm));
+               /*
+                * Some kernel threads have no associated pid.
+                * We still need to mark the entry as valid.
+                */
+               if (t->atts->pid)
+                       mapptr->valid = t->atts->pid;
+               else
+                       mapptr->valid = 1;
+
+               t->count++;
+       }
+}
+
+/*
+ *
+ * Writes a cpumap for the given iops_list/cpu_count to the provided buffer.
+ *
+ * You may provide a buffer and size, or if you set the buffer to NULL, a
+ * buffer of sufficient size will be allocated.
+ *
+ * If you provide a buffer and it is too small, sets cpumap_size to the number
+ * of bytes required and returns EINVAL.
+ *
+ * On success, if you provided a buffer, cpumap_size is set to the number of
+ * bytes written. If you did not provide a buffer, cpumap is set to the newly
+ * allocated buffer and cpumap_size is set to the number of bytes allocated.
+ *
+ * NOTE: It may seem redundant to pass both iops and a cpu_count.
+ *
+ * We may be reporting data from "now", or from the "past".
+ *
+ * The "now" data would be for something like kdbg_readcurcpumap().
+ * The "past" data would be for kdbg_readcpumap().
+ *
+ * If we do not pass both iops and cpu_count, and iops is NULL, this function
+ * will need to read "now" state to get the number of cpus, which would be in
+ * error if we were reporting "past" state.
+ */
+
+int
+kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, uint32_t* cpumap_size)
+{
+       assert(cpumap);
+       assert(cpumap_size);
+       assert(cpu_count);
+       assert(!iops || iops->cpu_id + 1 == cpu_count);
+
+       uint32_t bytes_needed = sizeof(kd_cpumap_header) + cpu_count * sizeof(kd_cpumap);
+       uint32_t bytes_available = *cpumap_size;
+       *cpumap_size = bytes_needed;
+       
+       if (*cpumap == NULL) {
+               if (kmem_alloc(kernel_map, (vm_offset_t*)cpumap, (vm_size_t)*cpumap_size) != KERN_SUCCESS) {
+                       return ENOMEM;
+               }
+       } else if (bytes_available < bytes_needed) {
+               return EINVAL;
+       }
+
+       kd_cpumap_header* header = (kd_cpumap_header*)(uintptr_t)*cpumap;
+
+       header->version_no = RAW_VERSION1;
+       header->cpu_count = cpu_count;
+
+       kd_cpumap* cpus = (kd_cpumap*)&header[1];
+
+       int32_t index = cpu_count - 1;
+       while (iops) {
+               cpus[index].cpu_id = iops->cpu_id;
+               cpus[index].flags = KDBG_CPUMAP_IS_IOP;
+               bzero(cpus[index].name, sizeof(cpus->name));
+               strlcpy(cpus[index].name, iops->callback.iop_name, sizeof(cpus->name));
+               
+               iops = iops->next;
+               index--;
+       }
+       
+       while (index >= 0) {
+               cpus[index].cpu_id = index;
+               cpus[index].flags = 0;
+               bzero(cpus[index].name, sizeof(cpus->name));
+               strlcpy(cpus[index].name, "AP", sizeof(cpus->name));
+
+               index--;
+       }
+       
+       return KERN_SUCCESS;
 }
 
 void
-kdbg_mapinit(void)
+kdbg_thrmap_init(void)
 {
-       struct proc *p;
-       struct krt akrt;
-       int tts_count;    /* number of task-to-string structures */
-       struct tts *tts_mapptr;
-       unsigned int tts_mapsize = 0;
-       unsigned int tts_maptomem=0;
-       int i;
+        if (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT)
+               return;
 
+       kd_mapptr = kdbg_thrmap_init_internal(0, &kd_mapsize, &kd_mapcount);
 
-        if (kdebug_flags & KDBG_MAPINIT)
-         return;
+       if (kd_mapptr)
+               kd_ctrl_page.kdebug_flags |= KDBG_MAPINIT;
+}
+
+
+kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, unsigned int *mapsize, unsigned int *mapcount)
+{
+       kd_threadmap    *mapptr;
+       struct proc     *p;
+       struct krt      akrt;
+       int             tts_count;    /* number of task-to-string structures */
+       struct tts      *tts_mapptr;
+       unsigned int    tts_mapsize = 0;
+       int             i;
+       vm_offset_t     kaddr;
 
-       /* Calculate the sizes of map buffers*/
-       for (p = allproc.lh_first, kd_mapcount=0, tts_count=0; p; 
-            p = p->p_list.le_next)
-         {
-           kd_mapcount += get_task_numacts((task_t)p->task);
-           tts_count++;
-         }
+       /*
+        * need to use PROC_SCANPROCLIST with proc_iterate
+        */
+       proc_list_lock();
+
+       /*
+        * Calculate the sizes of map buffers
+        */
+       for (p = allproc.lh_first, *mapcount=0, tts_count=0; p; p = p->p_list.le_next) {
+               *mapcount += get_task_numacts((task_t)p->task);
+               tts_count++;
+       }
+       proc_list_unlock();
 
        /*
         * The proc count could change during buffer allocation,
         * so introduce a small fudge factor to bump up the
         * buffer sizes. This gives new tasks some chance of 
-        * making into the tables.  Bump up by 10%.
+        * making into the tables.  Bump up by 25%.
         */
-       kd_mapcount += kd_mapcount/10;
-       tts_count += tts_count/10;
+       *mapcount += *mapcount/4;
+       tts_count += tts_count/4;
 
-       kd_mapsize = kd_mapcount * sizeof(kd_threadmap);
-       if((kmem_alloc(kernel_map, & kd_maptomem,
-                      (vm_size_t)kd_mapsize) == KERN_SUCCESS))
-       {
-           kd_mapptr = (kd_threadmap *) kd_maptomem;
-           bzero(kd_mapptr, kd_mapsize);
-       }
-       else
-           kd_mapptr = (kd_threadmap *) 0;
+       *mapsize = *mapcount * sizeof(kd_threadmap);
+
+       if (count && count < *mapcount)
+               return (0);
+
+       if ((kmem_alloc(kernel_map, &kaddr, (vm_size_t)*mapsize) == KERN_SUCCESS)) {
+               bzero((void *)kaddr, *mapsize);
+               mapptr = (kd_threadmap *)kaddr;
+       } else
+               return (0);
 
        tts_mapsize = tts_count * sizeof(struct tts);
-       if((kmem_alloc(kernel_map, & tts_maptomem,
-                      (vm_size_t)tts_mapsize) == KERN_SUCCESS))
-       {
-           tts_mapptr = (struct tts *) tts_maptomem;
-           bzero(tts_mapptr, tts_mapsize);
-       }
-       else
-           tts_mapptr = (struct tts *) 0;
 
+       if ((kmem_alloc(kernel_map, &kaddr, (vm_size_t)tts_mapsize) == KERN_SUCCESS)) {
+               bzero((void *)kaddr, tts_mapsize);
+               tts_mapptr = (struct tts *)kaddr;
+       } else {
+               kmem_free(kernel_map, (vm_offset_t)mapptr, *mapsize);
 
+               return (0);
+       }
        /* 
         * We need to save the procs command string
         * and take a reference for each task associated
         * with a valid process
         */
 
-       if (tts_mapptr) {
-               for (p = allproc.lh_first, i=0; p && i < tts_count; 
-                    p = p->p_list.le_next) {
-                       if (p->p_flag & P_WEXIT)
-                               continue;
-
-                       if (p->task) {
-                               task_reference(p->task);
-                               tts_mapptr[i].task = p->task;
-                               tts_mapptr[i].pid  = p->p_pid;
-                               (void)strncpy(tts_mapptr[i].task_comm, p->p_comm, sizeof(tts_mapptr[i].task_comm) - 1);
-                               i++;
-                       }
+       proc_list_lock();
+
+       /*
+        * should use proc_iterate
+        */
+       for (p = allproc.lh_first, i=0; p && i < tts_count; p = p->p_list.le_next) {
+               if (p->p_lflag & P_LEXIT)
+                       continue;
+
+               if (p->task) {
+                       task_reference(p->task);
+                       tts_mapptr[i].task = p->task;
+                       tts_mapptr[i].pid  = p->p_pid;
+                       (void)strlcpy(tts_mapptr[i].task_comm, p->p_comm, sizeof(tts_mapptr[i].task_comm));
+                       i++;
                }
-               tts_count = i;
        }
+       tts_count = i;
 
+       proc_list_unlock();
 
-       if (kd_mapptr && tts_mapptr)
-         {
-           kdebug_flags |= KDBG_MAPINIT;
-           /* Initialize thread map data */
-           akrt.map = kd_mapptr;
-           akrt.count = 0;
-           akrt.maxcount = kd_mapcount;
+       /*
+        * Initialize thread map data
+        */
+       akrt.map = mapptr;
+       akrt.count = 0;
+       akrt.maxcount = *mapcount;
            
-           for (i=0; i < tts_count; i++)
-             {
+       for (i = 0; i < tts_count; i++) {
                akrt.atts = &tts_mapptr[i];
                task_act_iterate_wth_args(tts_mapptr[i].task, kdbg_resolve_map, &akrt);
                task_deallocate((task_t) tts_mapptr[i].task);
-             }
-           kmem_free(kernel_map, (vm_offset_t)tts_mapptr, tts_mapsize);
-         }
+       }
+       kmem_free(kernel_map, (vm_offset_t)tts_mapptr, tts_mapsize);
+
+       *mapcount = akrt.count;
+
+       return (mapptr);
 }
 
 static void
@@ -654,9 +1655,7 @@ kdbg_clear(void)
         * First make sure we're not in
         * the middle of cutting a trace
         */
-
-       kdebug_enable &= ~KDEBUG_ENABLE_TRACE;
-       kdebug_slowcheck = SLOW_NOLOG;
+       kdbg_set_tracing_enabled(FALSE, KDEBUG_ENABLE_TRACE);
 
        /*
         * make sure the SLOW_NOLOG is seen
@@ -665,120 +1664,175 @@ kdbg_clear(void)
         */
        IOSleep(100);
 
-       if (kdebug_enable & KDEBUG_ENABLE_ENTROPY)
-               kdebug_slowcheck |= SLOW_ENTROPY;
-
         global_state_pid = -1;
-       kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-       kdebug_flags &= ~(KDBG_NOWRAP | KDBG_RANGECHECK | KDBG_VALCHECK);
-       kdebug_flags &= ~(KDBG_PIDCHECK | KDBG_PIDEXCLUDE);
+       kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+       kd_ctrl_page.kdebug_flags &= ~(KDBG_NOWRAP | KDBG_RANGECHECK | KDBG_VALCHECK);
+       kd_ctrl_page.kdebug_flags &= ~(KDBG_PIDCHECK | KDBG_PIDEXCLUDE);
+       
+       kdbg_disable_typefilter();
 
        delete_buffers();
+       nkdbufs = 0;
 
        /* Clean up the thread map buffer */
-       kdebug_flags &= ~KDBG_MAPINIT;
-       kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize);
-       kd_mapptr = (kd_threadmap *) 0;
+       kd_ctrl_page.kdebug_flags &= ~KDBG_MAPINIT;
+       if (kd_mapptr) {
+               kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize);
+               kd_mapptr = (kd_threadmap *) 0;
+       }
        kd_mapsize = 0;
        kd_mapcount = 0;
+
+       RAW_file_offset = 0;
+       RAW_file_written = 0;
 }
 
 int
 kdbg_setpid(kd_regtype *kdr)
 {
-  pid_t pid;
-  int flag, ret=0;
-  struct proc *p;
-
-  pid = (pid_t)kdr->value1;
-  flag = (int)kdr->value2;
-
-  if (pid > 0)
-    {
-      if ((p = pfind(pid)) == NULL)
-       ret = ESRCH;
-      else
-       {
-         if (flag == 1)  /* turn on pid check for this and all pids */
-           {
-             kdebug_flags |= KDBG_PIDCHECK;
-             kdebug_flags &= ~KDBG_PIDEXCLUDE;
-             kdebug_slowcheck |= SLOW_CHECKS;
-
-             p->p_flag |= P_KDEBUG;
-           }
-         else  /* turn off pid check for this pid value */
-           {
-             /* Don't turn off all pid checking though */
-             /* kdebug_flags &= ~KDBG_PIDCHECK;*/   
-             p->p_flag &= ~P_KDEBUG;
-           }
-       }
-    }
-  else
-    ret = EINVAL;
-  return(ret);
+       pid_t pid;
+       int flag, ret=0;
+       struct proc *p;
+
+       pid = (pid_t)kdr->value1;
+       flag = (int)kdr->value2;
+
+       if (pid > 0) {
+               if ((p = proc_find(pid)) == NULL)
+                       ret = ESRCH;
+               else {
+                       if (flag == 1) {
+                               /*
+                                * turn on pid check for this and all pids
+                                */
+                               kd_ctrl_page.kdebug_flags |= KDBG_PIDCHECK;
+                               kd_ctrl_page.kdebug_flags &= ~KDBG_PIDEXCLUDE;
+                               kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
+
+                               p->p_kdebug = 1;
+                       } else {
+                               /*
+                                * turn off pid check for this pid value
+                                * Don't turn off all pid checking though
+                                *
+                                * kd_ctrl_page.kdebug_flags &= ~KDBG_PIDCHECK;
+                                */   
+                               p->p_kdebug = 0;
+                       }
+                       proc_rele(p);
+               }
+       }
+       else
+               ret = EINVAL;
+
+       return(ret);
 }
 
 /* This is for pid exclusion in the trace buffer */
 int
 kdbg_setpidex(kd_regtype *kdr)
 {
-  pid_t pid;
-  int flag, ret=0;
-  struct proc *p;
-
-  pid = (pid_t)kdr->value1;
-  flag = (int)kdr->value2;
-
-  if (pid > 0)
-    {
-      if ((p = pfind(pid)) == NULL)
-       ret = ESRCH;
-      else
-       {
-         if (flag == 1)  /* turn on pid exclusion */
-           {
-             kdebug_flags |= KDBG_PIDEXCLUDE;
-             kdebug_flags &= ~KDBG_PIDCHECK;
-             kdebug_slowcheck |= SLOW_CHECKS;
-
-             p->p_flag |= P_KDEBUG;
-           }
-         else  /* turn off pid exclusion for this pid value */
-           {
-             /* Don't turn off all pid exclusion though */
-             /* kdebug_flags &= ~KDBG_PIDEXCLUDE;*/   
-             p->p_flag &= ~P_KDEBUG;
-           }
-       }
-    }
-  else
-    ret = EINVAL;
-  return(ret);
+       pid_t pid;
+       int flag, ret=0;
+       struct proc *p;
+
+       pid = (pid_t)kdr->value1;
+       flag = (int)kdr->value2;
+
+       if (pid > 0) {
+               if ((p = proc_find(pid)) == NULL)
+                       ret = ESRCH;
+               else {
+                       if (flag == 1) {
+                               /*
+                                * turn on pid exclusion
+                                */
+                               kd_ctrl_page.kdebug_flags |= KDBG_PIDEXCLUDE;
+                               kd_ctrl_page.kdebug_flags &= ~KDBG_PIDCHECK;
+                               kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
+
+                               p->p_kdebug = 1;
+                       }
+                       else {
+                               /*
+                                * turn off pid exclusion for this pid value
+                                * Don't turn off all pid exclusion though
+                                *
+                                * kd_ctrl_page.kdebug_flags &= ~KDBG_PIDEXCLUDE;
+                                */   
+                               p->p_kdebug = 0;
+                       }
+                       proc_rele(p);
+               }
+       } else
+               ret = EINVAL;
+
+       return(ret);
 }
 
-/* This is for setting a maximum decrementer value */
+
+/*
+ * This is for setting a maximum decrementer value
+ */
 int
 kdbg_setrtcdec(kd_regtype *kdr)
 {
-  int ret=0;
-  natural_t decval;
+       int ret = 0;
+       natural_t decval;
 
-  decval = (natural_t)kdr->value1;
+       decval = (natural_t)kdr->value1;
 
-  if (decval && decval < KDBG_MINRTCDEC)
-       ret = EINVAL;
-#ifdef ppc
-       else {
-               maxDec = decval ? decval : 0x7FFFFFFF;  /* Set or reset the max decrementer */
-       }
-#else
+       if (decval && decval < KDBG_MINRTCDEC)
+               ret = EINVAL;
        else
                ret = ENOTSUP;
-#endif /* ppc */
 
-  return(ret);
+       return(ret);
+}
+
+int
+kdbg_enable_typefilter(void)
+{
+       if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) {
+               /* free the old filter */
+               kdbg_disable_typefilter();
+       }
+       
+       if (kmem_alloc(kernel_map, (vm_offset_t *)&type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE) != KERN_SUCCESS) {
+               return ENOSPC;
+       }
+       
+       bzero(type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE);
+
+       /* Turn off range and value checks */
+       kd_ctrl_page.kdebug_flags &= ~(KDBG_RANGECHECK | KDBG_VALCHECK);
+       
+       /* Enable filter checking */
+       kd_ctrl_page.kdebug_flags |= KDBG_TYPEFILTER_CHECK;
+       kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
+       return 0;
+}
+
+int
+kdbg_disable_typefilter(void)
+{
+       /*  Disable filter checking */  
+       kd_ctrl_page.kdebug_flags &= ~KDBG_TYPEFILTER_CHECK;
+       
+       /*  Turn off slow checks unless pid checks are using them */
+       if ( (kd_ctrl_page.kdebug_flags & (KDBG_PIDCHECK | KDBG_PIDEXCLUDE)) )
+               kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
+       else
+               kdbg_set_flags(SLOW_CHECKS, 0, FALSE);
+       
+       if(type_filter_bitmap == NULL)
+               return 0;
+
+       vm_offset_t old_bitmap = (vm_offset_t)type_filter_bitmap;
+       type_filter_bitmap = NULL;
+
+       kmem_free(kernel_map, old_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE);
+       return 0;
 }
 
 int
@@ -793,10 +1847,10 @@ kdbg_setreg(kd_regtype * kdr)
                val_2 = (kdr->value2 & 0xff);
                kdlog_beg = (val_1<<24);
                kdlog_end = (val_2<<24);
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-               kdebug_flags &= ~KDBG_VALCHECK;       /* Turn off specific value check  */
-               kdebug_flags |= (KDBG_RANGECHECK | KDBG_CLASSTYPE);
-               kdebug_slowcheck |= SLOW_CHECKS;
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags &= ~KDBG_VALCHECK;       /* Turn off specific value check  */
+               kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_CLASSTYPE);
+               kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
                break;
        case KDBG_SUBCLSTYPE :
                val_1 = (kdr->value1 & 0xff);
@@ -804,36 +1858,38 @@ kdbg_setreg(kd_regtype * kdr)
                val = val_2 + 1;
                kdlog_beg = ((val_1<<24) | (val_2 << 16));
                kdlog_end = ((val_1<<24) | (val << 16));
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-               kdebug_flags &= ~KDBG_VALCHECK;       /* Turn off specific value check  */
-               kdebug_flags |= (KDBG_RANGECHECK | KDBG_SUBCLSTYPE);
-               kdebug_slowcheck |= SLOW_CHECKS;
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags &= ~KDBG_VALCHECK;       /* Turn off specific value check  */
+               kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_SUBCLSTYPE);
+               kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
                break;
        case KDBG_RANGETYPE :
                kdlog_beg = (kdr->value1);
                kdlog_end = (kdr->value2);
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-               kdebug_flags &= ~KDBG_VALCHECK;       /* Turn off specific value check  */
-               kdebug_flags |= (KDBG_RANGECHECK | KDBG_RANGETYPE);
-               kdebug_slowcheck |= SLOW_CHECKS;
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags &= ~KDBG_VALCHECK;       /* Turn off specific value check  */
+               kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_RANGETYPE);
+               kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
                break;
        case KDBG_VALCHECK:
                kdlog_value1 = (kdr->value1);
                kdlog_value2 = (kdr->value2);
                kdlog_value3 = (kdr->value3);
                kdlog_value4 = (kdr->value4);
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-               kdebug_flags &= ~KDBG_RANGECHECK;    /* Turn off range check */
-               kdebug_flags |= KDBG_VALCHECK;       /* Turn on specific value check  */
-               kdebug_slowcheck |= SLOW_CHECKS;
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags &= ~KDBG_RANGECHECK;    /* Turn off range check */
+               kd_ctrl_page.kdebug_flags |= KDBG_VALCHECK;       /* Turn on specific value check  */
+               kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
                break;
        case KDBG_TYPENONE :
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
 
-               if ( (kdebug_flags & (KDBG_RANGECHECK | KDBG_VALCHECK | KDBG_PIDCHECK | KDBG_PIDEXCLUDE)) )
-                       kdebug_slowcheck |= SLOW_CHECKS;
+               if ( (kd_ctrl_page.kdebug_flags & (KDBG_RANGECHECK | KDBG_VALCHECK   | 
+                                                  KDBG_PIDCHECK   | KDBG_PIDEXCLUDE | 
+                                                  KDBG_TYPEFILTER_CHECK)) )
+                       kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
                else
-                       kdebug_slowcheck &= ~SLOW_CHECKS;
+                       kdbg_set_flags(SLOW_CHECKS, 0, FALSE);
 
                kdlog_beg = 0;
                kdlog_end = 0;
@@ -858,8 +1914,8 @@ kdbg_getreg(__unused kd_regtype * kdr)
                val_2 = val_1 + 1;
                kdlog_beg = (val_1<<24);
                kdlog_end = (val_2<<24);
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-               kdebug_flags |= (KDBG_RANGECHECK | KDBG_CLASSTYPE);
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_CLASSTYPE);
                break;
        case KDBG_SUBCLSTYPE :
                val_1 = (kdr->value1 & 0xff);
@@ -867,17 +1923,17 @@ kdbg_getreg(__unused kd_regtype * kdr)
                val = val_2 + 1;
                kdlog_beg = ((val_1<<24) | (val_2 << 16));
                kdlog_end = ((val_1<<24) | (val << 16));
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-               kdebug_flags |= (KDBG_RANGECHECK | KDBG_SUBCLSTYPE);
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_SUBCLSTYPE);
                break;
        case KDBG_RANGETYPE :
                kdlog_beg = (kdr->value1);
                kdlog_end = (kdr->value2);
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
-               kdebug_flags |= (KDBG_RANGECHECK | KDBG_RANGETYPE);
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_RANGETYPE);
                break;
        case KDBG_TYPENONE :
-               kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
+               kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
                kdlog_beg = 0;
                kdlog_end = 0;
                break;
@@ -889,110 +1945,240 @@ kdbg_getreg(__unused kd_regtype * kdr)
        return(EINVAL);
 }
 
+int
+kdbg_readcpumap(user_addr_t user_cpumap, size_t *user_cpumap_size)
+{
+       uint8_t* cpumap = NULL;
+       uint32_t cpumap_size = 0;
+       int ret = KERN_SUCCESS;
+
+       if (kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) {
+               if (kdbg_cpumap_init_internal(kd_ctrl_page.kdebug_iops, kd_ctrl_page.kdebug_cpus, &cpumap, &cpumap_size) == KERN_SUCCESS) {
+                       if (user_cpumap) {
+                               size_t bytes_to_copy = (*user_cpumap_size >= cpumap_size) ? cpumap_size : *user_cpumap_size;
+                               if (copyout(cpumap, user_cpumap, (size_t)bytes_to_copy)) {
+                                       ret = EFAULT;
+                               }
+                       }
+                       *user_cpumap_size = cpumap_size;
+                       kmem_free(kernel_map, (vm_offset_t)cpumap, cpumap_size);
+               } else
+                       ret = EINVAL;
+       } else
+               ret = EINVAL;
+
+       return (ret);
+}
 
 int
-kdbg_readmap(user_addr_t buffer, size_t *number)
+kdbg_readcurthrmap(user_addr_t buffer, size_t *bufsize)
 {
-  int avail = *number;
-  int ret = 0;
-  unsigned int count = 0;
+       kd_threadmap *mapptr;
+       unsigned int mapsize;
+       unsigned int mapcount;
+       unsigned int count = 0;
+       int ret = 0;
+
+       count = *bufsize/sizeof(kd_threadmap);
+       *bufsize = 0;
+
+       if ( (mapptr = kdbg_thrmap_init_internal(count, &mapsize, &mapcount)) ) {
+               if (copyout(mapptr, buffer, mapcount * sizeof(kd_threadmap)))
+                       ret = EFAULT;
+               else
+                       *bufsize = (mapcount * sizeof(kd_threadmap));
 
-  count = avail/sizeof (kd_threadmap);
+               kmem_free(kernel_map, (vm_offset_t)mapptr, mapsize);
+       } else
+               ret = EINVAL;
 
-  if (count && (count <= kd_mapcount))
-    {
-      if((kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr)
-       {
-         if (*number < kd_mapsize)
-           ret=EINVAL;
-         else
-           {
-             if (copyout(kd_mapptr, buffer, kd_mapsize))
-               ret=EINVAL;
-           }
-       }
-      else
-       ret=EINVAL;
-    }
-  else
-    ret=EINVAL;
-
-  if ((kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr)
-    {
-      kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize);
-      kdebug_flags &= ~KDBG_MAPINIT;
-      kd_mapsize = 0;
-      kd_mapptr = (kd_threadmap *) 0;
-      kd_mapcount = 0;
-    }  
-
-  return(ret);
+       return (ret);
 }
 
 int
-kdbg_getentropy (user_addr_t buffer, size_t *number, int ms_timeout)
+kdbg_readthrmap(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx)
+{
+       int avail = *number;
+       int ret = 0;
+       uint32_t count = 0;
+       unsigned int mapsize;
+
+       count = avail/sizeof (kd_threadmap);
+
+       mapsize = kd_mapcount * sizeof(kd_threadmap);
+
+       if (count && (count <= kd_mapcount))
+       {
+               if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr)
+               {
+                       if (*number < mapsize)
+                               ret = EINVAL;
+                       else
+                       {
+                               if (vp)
+                               {
+                                       RAW_header      header;
+                                       clock_sec_t     secs;
+                                       clock_usec_t    usecs;
+                                       char    *pad_buf;
+                                       uint32_t pad_size;
+                                       uint32_t extra_thread_count = 0;
+                                       uint32_t cpumap_size;
+                                       
+                                       /*
+                                        * To write a RAW_VERSION1+ file, we
+                                        * must embed a cpumap in the "padding"
+                                        * used to page align the events folloing
+                                        * the threadmap. If the threadmap happens
+                                        * to not require enough padding, we
+                                        * artificially increase its footprint
+                                        * until it needs enough padding.
+                                        */
+
+                                       pad_size = PAGE_SIZE - ((sizeof(RAW_header) + (count * sizeof(kd_threadmap))) & PAGE_MASK_64);
+                                       cpumap_size = sizeof(kd_cpumap_header) + kd_ctrl_page.kdebug_cpus * sizeof(kd_cpumap);
+
+                                       if (cpumap_size > pad_size) {
+                                               /* Force an overflow onto the next page, we get a full page of padding */
+                                               extra_thread_count = (pad_size / sizeof(kd_threadmap)) + 1;
+                                       }
+
+                                       header.version_no = RAW_VERSION1;
+                                       header.thread_count = count + extra_thread_count;
+
+                                       clock_get_calendar_microtime(&secs, &usecs);
+                                       header.TOD_secs = secs;
+                                       header.TOD_usecs = usecs;
+
+                                       ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)&header, sizeof(RAW_header), RAW_file_offset,
+                                                     UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
+                                       if (ret)
+                                               goto write_error;
+                                       RAW_file_offset += sizeof(RAW_header);
+
+                                       ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)kd_mapptr, mapsize, RAW_file_offset,
+                                                     UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
+                                       if (ret)
+                                               goto write_error;
+                                       RAW_file_offset += mapsize;
+
+                                       if (extra_thread_count) {
+                                               pad_size = extra_thread_count * sizeof(kd_threadmap);
+                                               pad_buf = (char *)kalloc(pad_size);
+                                               memset(pad_buf, 0, pad_size);
+
+                                               ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)pad_buf, pad_size, RAW_file_offset,
+                                                             UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
+                                               kfree(pad_buf, pad_size);
+
+                                               if (ret)
+                                                       goto write_error;
+                                               RAW_file_offset += pad_size;
+
+                                       }
+                                       
+                                       pad_size = PAGE_SIZE - (RAW_file_offset & PAGE_MASK_64);
+                                       if (pad_size) {
+                                               pad_buf = (char *)kalloc(pad_size);
+                                               memset(pad_buf, 0, pad_size);
+
+                                               /*
+                                                * embed a cpumap in the padding bytes.
+                                                * older code will skip this.
+                                                * newer code will know how to read it.
+                                                */
+                                               uint32_t temp = pad_size;
+                                               if (kdbg_cpumap_init_internal(kd_ctrl_page.kdebug_iops, kd_ctrl_page.kdebug_cpus, (uint8_t**)&pad_buf, &temp) != KERN_SUCCESS) {
+                                                       memset(pad_buf, 0, pad_size);
+                                               }
+
+                                               ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)pad_buf, pad_size, RAW_file_offset,
+                                                             UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
+                                               kfree(pad_buf, pad_size);
+
+                                               if (ret)
+                                                       goto write_error;
+                                               RAW_file_offset += pad_size;
+                                       }
+                                       RAW_file_written += sizeof(RAW_header) + mapsize + pad_size;
+
+                               } else {
+                                       if (copyout(kd_mapptr, buffer, mapsize))
+                                               ret = EINVAL;
+                               }
+                       }
+               }
+               else
+                       ret = EINVAL;
+       }
+       else
+               ret = EINVAL;
+
+       if (ret && vp)
+       {
+               count = 0;
+
+               vn_rdwr(UIO_WRITE, vp, (caddr_t)&count, sizeof(uint32_t), RAW_file_offset,
+                       UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
+               RAW_file_offset += sizeof(uint32_t);
+               RAW_file_written += sizeof(uint32_t);
+       }
+write_error:
+       if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr)
+       {
+               kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize);
+               kd_ctrl_page.kdebug_flags &= ~KDBG_MAPINIT;
+               kd_mapsize = 0;
+               kd_mapptr = (kd_threadmap *) 0;
+               kd_mapcount = 0;
+       }  
+       return(ret);
+}
+
+
+static int
+kdbg_set_nkdbufs(unsigned int value)
+{
+        /*
+        * We allow a maximum buffer size of 50% of either ram or max mapped address, whichever is smaller
+        * 'value' is the desired number of trace entries
+        */
+        unsigned int max_entries = (sane_size/2) / sizeof(kd_buf);
+
+       if (value <= max_entries)
+               return (value);
+       else
+               return (max_entries);
+}
+
+
+static int
+kdbg_enable_bg_trace(void)
+{
+       int ret = 0;
+
+       if (kdlog_bg_trace == TRUE && kdlog_bg_trace_running == FALSE && n_storage_buffers == 0) {
+               nkdbufs = bg_nkdbufs;
+               ret = kdbg_reinit(FALSE);
+               if (0 == ret) {
+                       kdbg_set_tracing_enabled(TRUE, KDEBUG_ENABLE_TRACE);
+                       kdlog_bg_trace_running = TRUE;
+               }
+       }
+       return ret;
+}
+
+static void
+kdbg_disable_bg_trace(void)
 {
-  int avail = *number;
-  int ret = 0;
-
-  if (kd_entropy_buffer)
-    return(EBUSY);
-
-  kd_entropy_count = avail/sizeof(mach_timespec_t);
-  kd_entropy_bufsize = kd_entropy_count * sizeof(mach_timespec_t);
-  kd_entropy_indx = 0;
-
-  /* Enforce maximum entropy entries here if needed */
-
-  /* allocate entropy buffer */
-  if (kmem_alloc(kernel_map, &kd_entropy_buftomem,
-                (vm_size_t)kd_entropy_bufsize) == KERN_SUCCESS)
-    {
-      kd_entropy_buffer = (uint64_t *) kd_entropy_buftomem;
-    }
-  else
-    {
-      kd_entropy_buffer = (uint64_t *) 0;
-      kd_entropy_count = 0;
-      kd_entropy_indx = 0;
-      return (EINVAL);
-    }
-
-  if (ms_timeout < 10)
-    ms_timeout = 10;
-
-  /* Enable entropy sampling */
-  kdebug_enable |= KDEBUG_ENABLE_ENTROPY;
-  kdebug_slowcheck |= SLOW_ENTROPY;
-
-  ret = tsleep (kdbg_getentropy, PRIBIO | PCATCH, "kd_entropy", (ms_timeout/(1000/HZ)));
-
-  /* Disable entropy sampling */
-  kdebug_enable &= ~KDEBUG_ENABLE_ENTROPY;
-  kdebug_slowcheck &= ~SLOW_ENTROPY;
-
-  *number = 0;
-  ret = 0;
-
-  if (kd_entropy_indx > 0)
-    {
-      /* copyout the buffer */
-      if (copyout(kd_entropy_buffer, buffer, kd_entropy_indx * sizeof(mach_timespec_t)))
-         ret = EINVAL;
-      else
-         *number = kd_entropy_indx;
-    }
-
-  /* Always cleanup */
-  kd_entropy_count = 0;
-  kd_entropy_indx = 0;
-  kd_entropy_buftomem = 0;
-  kmem_free(kernel_map, (vm_offset_t)kd_entropy_buffer, kd_entropy_bufsize);
-  kd_entropy_buffer = (uint64_t *) 0;
-  return(ret);
+       if (kdlog_bg_trace_running == TRUE) {
+               kdlog_bg_trace_running = FALSE;
+               kdbg_clear();           
+       }
 }
 
 
+
 /*
  * This function is provided for the CHUD toolkit only.
  *    int val:
@@ -1005,199 +2191,372 @@ kdbg_getentropy (user_addr_t buffer, size_t *number, int ms_timeout)
 void
 kdbg_control_chud(int val, void *fn)
 {
-        if (val) {
-                /* enable chudhook */
+       kdbg_lock_init();
+    
+       if (val) {
+               /* enable chudhook */
                kdebug_chudhook = fn;
-               kdebug_enable |= KDEBUG_ENABLE_CHUD;
+               kdbg_set_flags(SLOW_CHUD, KDEBUG_ENABLE_CHUD, TRUE);
        }
        else {
-               /* disable chudhook */
-                kdebug_enable &= ~KDEBUG_ENABLE_CHUD;
+               /* disable chudhook */
+               kdbg_set_flags(SLOW_CHUD, KDEBUG_ENABLE_CHUD, FALSE);
                kdebug_chudhook = 0;
        }
 }
 
        
 int
-kdbg_control(int *name, __unused u_int namelen, user_addr_t where, size_t *sizep)
+kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep)
 {
-    int ret=0;
-       size_t size=*sizep;
-       unsigned int max_entries;
-       unsigned int value = name[1];
+       int ret = 0;
+       size_t size = *sizep;
+       unsigned int value = 0;
        kd_regtype kd_Reg;
        kbufinfo_t kd_bufinfo;
        pid_t curpid;
-       struct proc *p, *curproc;
-
-
+       proc_t p, curproc;
+
+       if (name[0] == KERN_KDGETENTROPY ||
+               name[0] == KERN_KDWRITETR ||
+               name[0] == KERN_KDWRITEMAP ||
+               name[0] == KERN_KDEFLAGS ||
+               name[0] == KERN_KDDFLAGS ||
+               name[0] == KERN_KDENABLE ||
+               name[0] == KERN_KDENABLE_BG_TRACE ||
+               name[0] == KERN_KDSETBUF) {
+               
+               if ( namelen < 2 )
+                       return(EINVAL);
+               value = name[1];
+       }
+       
        kdbg_lock_init();
 
-       if ( !(kdebug_flags & KDBG_LOCKINIT))
-               return(ENOMEM);
+       if ( !(kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT))
+               return(ENOSPC);
 
        lck_mtx_lock(kd_trace_mtx_sysctl);
 
-       if (name[0] == KERN_KDGETBUF) {
-           /* 
-            * Does not alter the global_state_pid
-            * This is a passive request.
-            */
-           if (size < sizeof(kd_bufinfo.nkdbufs)) {
-               /* 
-                * There is not enough room to return even
-                * the first element of the info structure.
-                */
-               lck_mtx_unlock(kd_trace_mtx_sysctl);
-
-               return(EINVAL);
-           }
-           kd_bufinfo.nkdbufs = nkdbufs;
-           kd_bufinfo.nkdthreads = kd_mapsize / sizeof(kd_threadmap);
-
-           if ( (kdebug_slowcheck & SLOW_NOLOG) )
-               kd_bufinfo.nolog = 1;
-           else
-               kd_bufinfo.nolog = 0;
-           kd_bufinfo.flags = kdebug_flags;
-           kd_bufinfo.bufid = global_state_pid;
-          
-           if (size >= sizeof(kd_bufinfo)) {
-            /*
-                    * Provide all the info we have
-                   */
-               if (copyout (&kd_bufinfo, where, sizeof(kd_bufinfo))) {
-                       lck_mtx_unlock(kd_trace_mtx_sysctl);
-
-                return(EINVAL);
-                   }
-           }
-           else {
-               /* 
-                        * For backwards compatibility, only provide
-                        * as much info as there is room for.
-                        */
-               if (copyout (&kd_bufinfo, where, size)) {
-                       lck_mtx_unlock(kd_trace_mtx_sysctl);
-
-                       return(EINVAL);
-                   }
-           }
-           lck_mtx_unlock(kd_trace_mtx_sysctl);
-
-           return(0);
-       } else if (name[0] == KERN_KDGETENTROPY) {
-           if (kd_entropy_buffer)
-               ret = EBUSY;
-           else
-               ret = kdbg_getentropy(where, sizep, value);
-           lck_mtx_unlock(kd_trace_mtx_sysctl);
-
-           return (ret);
+       switch(name[0]) {
+               case KERN_KDGETBUF:
+                       /*
+                        * Does not alter the global_state_pid
+                        * This is a passive request.
+                        */
+                       if (size < sizeof(kd_bufinfo.nkdbufs)) {
+                               /*
+                                * There is not enough room to return even
+                                * the first element of the info structure.
+                                */
+                               ret = EINVAL;
+                               goto out;
+                       }
+                       kd_bufinfo.nkdbufs = nkdbufs;
+                       kd_bufinfo.nkdthreads = kd_mapcount;
+                       
+                       if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) )
+                               kd_bufinfo.nolog = 1;
+                       else
+                               kd_bufinfo.nolog = 0;
+                       
+                       kd_bufinfo.flags = kd_ctrl_page.kdebug_flags;
+#if defined(__LP64__)
+                       kd_bufinfo.flags |= KDBG_LP64;
+#endif
+                       kd_bufinfo.bufid = global_state_pid;
+                       
+                       if (size >= sizeof(kd_bufinfo)) {
+                               /*
+                                * Provide all the info we have
+                                */
+                               if (copyout(&kd_bufinfo, where, sizeof(kd_bufinfo)))
+                                       ret = EINVAL;
+                       } else {
+                               /*
+                                * For backwards compatibility, only provide
+                                * as much info as there is room for.
+                                */
+                               if (copyout(&kd_bufinfo, where, size))
+                                       ret = EINVAL;
+                       }
+                       goto out;
+                       
+               case KERN_KDGETENTROPY: {
+                       /* Obsolescent - just fake with a random buffer */
+                       char    *buffer = (char *) kalloc(size);
+                       read_frandom((void *) buffer, size);
+                       ret = copyout(buffer, where, size);
+                       kfree(buffer, size);
+                       goto out;
+               }
+                       
+               case KERN_KDENABLE_BG_TRACE:
+                       bg_nkdbufs = kdbg_set_nkdbufs(value);
+                       kdlog_bg_trace = TRUE;
+                       ret = kdbg_enable_bg_trace();
+                       goto out;
+                       
+               case KERN_KDDISABLE_BG_TRACE:
+                       kdlog_bg_trace = FALSE;
+                       kdbg_disable_bg_trace();
+                       goto out;
        }
        
        if ((curproc = current_proc()) != NULL)
-           curpid = curproc->p_pid;
+               curpid = curproc->p_pid;
        else {
-           lck_mtx_unlock(kd_trace_mtx_sysctl);
-
-           return (ESRCH);
+               ret = ESRCH;
+               goto out;
        }
-        if (global_state_pid == -1)
-           global_state_pid = curpid;
+       if (global_state_pid == -1)
+               global_state_pid = curpid;
        else if (global_state_pid != curpid) {
-           if ((p = pfind(global_state_pid)) == NULL) {
-               /*
-                * The global pid no longer exists
-                */
-               global_state_pid = curpid;
-           } else {
-               /*
-                * The global pid exists, deny this request
-                */
-               lck_mtx_unlock(kd_trace_mtx_sysctl);
+               if ((p = proc_find(global_state_pid)) == NULL) {
+                       /*
+                        * The global pid no longer exists
+                        */
+                       global_state_pid = curpid;
+               } else {
+                       /*
+                        * The global pid exists, deny this request
+                        */
+                       proc_rele(p);
 
-               return(EBUSY);
-           }
+                       ret = EBUSY;
+                       goto out;
+               }
        }
 
        switch(name[0]) {
                case KERN_KDEFLAGS:
+                       kdbg_disable_bg_trace();
+
                        value &= KDBG_USERFLAGS;
-                       kdebug_flags |= value;
+                       kd_ctrl_page.kdebug_flags |= value;
                        break;
                case KERN_KDDFLAGS:
+                       kdbg_disable_bg_trace();
+
                        value &= KDBG_USERFLAGS;
-                       kdebug_flags &= ~value;
+                       kd_ctrl_page.kdebug_flags &= ~value;
                        break;
-               case KERN_KDENABLE:    /* used to enable or disable */
-                 if (value)
-                   {
-                     /* enable only if buffer is initialized */
-                     if (!(kdebug_flags & KDBG_BUFINIT))
+               case KERN_KDENABLE:
+                       /*
+                        * Enable tracing mechanism.  Two types:
+                        * KDEBUG_TRACE is the standard one,
+                        * and KDEBUG_PPT which is a carefully
+                        * chosen subset to avoid performance impact.
+                        */
+                       if (value) {
+                               /*
+                                * enable only if buffer is initialized
+                                */
+                               if (!(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) || 
+                                   !(value == KDEBUG_ENABLE_TRACE || value == KDEBUG_ENABLE_PPT)) {
+                                       ret = EINVAL;
+                                       break;
+                               }
+                               kdbg_thrmap_init();
+
+                               kdbg_set_tracing_enabled(TRUE, value);
+                       }
+                       else
                        {
-                         ret=EINVAL;
-                         break;
+                               kdbg_set_tracing_enabled(FALSE, 0);
                        }
-                     kdbg_mapinit();
-
-                     kdebug_enable |= KDEBUG_ENABLE_TRACE;
-                     kdebug_slowcheck &= ~SLOW_NOLOG;
-                   }
-                 else
-                   {
-                     kdebug_enable &= ~KDEBUG_ENABLE_TRACE;
-                     kdebug_slowcheck |= SLOW_NOLOG;
-                   }
-                 break;
+                       break;
                case KERN_KDSETBUF:
-                 /* We allow a maximum buffer size of 25% of either ram or max mapped address, whichever is smaller */
-                 /* 'value' is the desired number of trace entries */
-                       max_entries = (sane_size/4) / sizeof(kd_buf);
-                       if (value <= max_entries)
-                               nkdbufs = value;
-                       else
-                         nkdbufs = max_entries;
+                       kdbg_disable_bg_trace();
+
+                       nkdbufs = kdbg_set_nkdbufs(value);
                        break;
                case KERN_KDSETUP:
-                       ret=kdbg_reinit();
+                       kdbg_disable_bg_trace();
+
+                       ret = kdbg_reinit(FALSE);
                        break;
                case KERN_KDREMOVE:
                        kdbg_clear();
+                       ret = kdbg_enable_bg_trace();
                        break;
                case KERN_KDSETREG:
                        if(size < sizeof(kd_regtype)) {
-                               ret=EINVAL;
+                               ret = EINVAL;
                                break;
                        }
                        if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
-                               ret= EINVAL;
+                               ret = EINVAL;
                                break;
                        }
+                       kdbg_disable_bg_trace();
+
                        ret = kdbg_setreg(&kd_Reg);
                        break;
                case KERN_KDGETREG:
-                       if(size < sizeof(kd_regtype)) {
+                       if (size < sizeof(kd_regtype)) {
                                ret = EINVAL;
                                break;
                        }
                        ret = kdbg_getreg(&kd_Reg);
-                       if (copyout(&kd_Reg, where, sizeof(kd_regtype))){
-                               ret=EINVAL;
+                       if (copyout(&kd_Reg, where, sizeof(kd_regtype))) {
+                               ret = EINVAL;
                        }
+                       kdbg_disable_bg_trace();
+
                        break;
                case KERN_KDREADTR:
-                       ret = kdbg_read(where, sizep);
+                       ret = kdbg_read(where, sizep, NULL, NULL);
+                       break;
+               case KERN_KDWRITETR:
+               case KERN_KDWRITEMAP:
+               {
+                       struct  vfs_context context;
+                       struct  fileproc *fp;
+                       size_t  number;
+                       vnode_t vp;
+                       int     fd;
+
+                       kdbg_disable_bg_trace();
+
+                       if (name[0] == KERN_KDWRITETR) {
+                               int s;
+                               int wait_result = THREAD_AWAKENED;
+                               u_int64_t abstime;
+                               u_int64_t ns;
+
+                               if (*sizep) {
+                                       ns = ((u_int64_t)*sizep) * (u_int64_t)(1000 * 1000);
+                                       nanoseconds_to_absolutetime(ns,  &abstime );
+                                       clock_absolutetime_interval_to_deadline( abstime, &abstime );
+                               } else
+                                       abstime = 0;
+
+                               s = ml_set_interrupts_enabled(FALSE);
+                               lck_spin_lock(kdw_spin_lock);
+
+                               while (wait_result == THREAD_AWAKENED && kd_ctrl_page.kds_inuse_count < n_storage_threshold) {
+
+                                       kds_waiter = 1;
+
+                                       if (abstime)
+                                               wait_result = lck_spin_sleep_deadline(kdw_spin_lock, 0, &kds_waiter, THREAD_ABORTSAFE, abstime);
+                                       else
+                                               wait_result = lck_spin_sleep(kdw_spin_lock, 0, &kds_waiter, THREAD_ABORTSAFE);
+                                       
+                                       kds_waiter = 0;
+                               }
+                               lck_spin_unlock(kdw_spin_lock);
+                               ml_set_interrupts_enabled(s);
+                       }
+                       p = current_proc();
+                       fd = value;
+
+                       proc_fdlock(p);
+                       if ( (ret = fp_lookup(p, fd, &fp, 1)) ) {
+                               proc_fdunlock(p);
+                               break;
+                       }
+                       context.vc_thread = current_thread();
+                       context.vc_ucred = fp->f_fglob->fg_cred;
+
+                       if (FILEGLOB_DTYPE(fp->f_fglob) != DTYPE_VNODE) {
+                               fp_drop(p, fd, fp, 1);
+                               proc_fdunlock(p);
+
+                               ret = EBADF;
+                               break;
+                       }
+                       vp = (struct vnode *)fp->f_fglob->fg_data;
+                       proc_fdunlock(p);
+
+                       if ((ret = vnode_getwithref(vp)) == 0) {
+                               RAW_file_offset = fp->f_fglob->fg_offset;
+                               if (name[0] == KERN_KDWRITETR) {
+                                       number = nkdbufs * sizeof(kd_buf);
+
+                                       KERNEL_DEBUG_CONSTANT(TRACE_WRITING_EVENTS | DBG_FUNC_START, 0, 0, 0, 0, 0);
+                                       ret = kdbg_read(0, &number, vp, &context);
+                                       KERNEL_DEBUG_CONSTANT(TRACE_WRITING_EVENTS | DBG_FUNC_END, number, 0, 0, 0, 0);
+
+                                       *sizep = number;
+                               } else {
+                                       number = kd_mapcount * sizeof(kd_threadmap);
+                                       kdbg_readthrmap(0, &number, vp, &context);
+                               }
+                               fp->f_fglob->fg_offset = RAW_file_offset;
+                               vnode_put(vp);
+                       }
+                       fp_drop(p, fd, fp, 0);
+
                        break;
+               }
+               case KERN_KDBUFWAIT:
+               {
+                       /* WRITETR lite -- just block until there's data */
+                       int s;
+                       int wait_result = THREAD_AWAKENED;
+                       u_int64_t abstime;
+                       u_int64_t ns;
+                       size_t  number = 0;
+
+                       kdbg_disable_bg_trace();
+
+
+                       if (*sizep) {
+                               ns = ((u_int64_t)*sizep) * (u_int64_t)(1000 * 1000);
+                               nanoseconds_to_absolutetime(ns,  &abstime );
+                               clock_absolutetime_interval_to_deadline( abstime, &abstime );
+                       } else
+                               abstime = 0;
+
+                       s = ml_set_interrupts_enabled(FALSE);
+                       if( !s )
+                               panic("trying to wait with interrupts off");
+                       lck_spin_lock(kdw_spin_lock);
+
+                       /* drop the mutex so don't exclude others from
+                        * accessing trace
+                        */
+                       lck_mtx_unlock(kd_trace_mtx_sysctl);
+
+                       while (wait_result == THREAD_AWAKENED &&
+                               kd_ctrl_page.kds_inuse_count < n_storage_threshold) {
+
+                               kds_waiter = 1;
+
+                               if (abstime)
+                                       wait_result = lck_spin_sleep_deadline(kdw_spin_lock, 0, &kds_waiter, THREAD_ABORTSAFE, abstime);
+                               else
+                                       wait_result = lck_spin_sleep(kdw_spin_lock, 0, &kds_waiter, THREAD_ABORTSAFE);
+                               
+                               kds_waiter = 0;
+                       }
+
+                       /* check the count under the spinlock */
+                       number = (kd_ctrl_page.kds_inuse_count >= n_storage_threshold);
+
+                       lck_spin_unlock(kdw_spin_lock);
+                       ml_set_interrupts_enabled(s);
+
+                       /* pick the mutex back up again */
+                       lck_mtx_lock(kd_trace_mtx_sysctl);
+
+                       /* write out whether we've exceeded the threshold */
+                       *sizep = number;
+                       break;
+               }
                case KERN_KDPIDTR:
                        if (size < sizeof(kd_regtype)) {
                                ret = EINVAL;
                                break;
                        }
                        if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
-                               ret= EINVAL;
+                               ret = EINVAL;
                                break;
                        }
+                       kdbg_disable_bg_trace();
+
                        ret = kdbg_setpid(&kd_Reg);
                        break;
                case KERN_KDPIDEX:
@@ -1206,29 +2565,58 @@ kdbg_control(int *name, __unused u_int namelen, user_addr_t where, size_t *sizep
                                break;
                        }
                        if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
-                               ret= EINVAL;
+                               ret = EINVAL;
                                break;
                        }
+                       kdbg_disable_bg_trace();
+
                        ret = kdbg_setpidex(&kd_Reg);
                        break;
-               case KERN_KDTHRMAP:
-                       ret = kdbg_readmap(where, sizep);
-                       break;
-               case KERN_KDSETRTCDEC:
+               case KERN_KDCPUMAP:
+                       ret = kdbg_readcpumap(where, sizep);
+                       break;
+               case KERN_KDTHRMAP:
+                       ret = kdbg_readthrmap(where, sizep, NULL, NULL);
+                       break;
+               case KERN_KDREADCURTHRMAP:
+                       ret = kdbg_readcurthrmap(where, sizep);
+                       break;
+               case KERN_KDSETRTCDEC:
                        if (size < sizeof(kd_regtype)) {
                                ret = EINVAL;
                                break;
                        }
                        if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
-                               ret= EINVAL;
+                               ret = EINVAL;
                                break;
                        }
+                       kdbg_disable_bg_trace();
+
                        ret = kdbg_setrtcdec(&kd_Reg);
                        break;
-                      
+               case KERN_KDSET_TYPEFILTER:
+                       kdbg_disable_bg_trace();
+
+                       if ((kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) == 0){
+                               if ((ret = kdbg_enable_typefilter()))
+                                       break;
+                       }
+
+                       if (size != KDBG_TYPEFILTER_BITMAP_SIZE) {
+                               ret = EINVAL;
+                               break;
+                       }
+
+                       if (copyin(where, type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE)) {
+                               ret = EINVAL;
+                               break;
+                       }
+                       kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_TYPEFILTER_CHANGED, type_filter_bitmap);
+                       break;
                default:
-                       ret= EINVAL;
+                       ret = EINVAL;
        }
+out:
        lck_mtx_unlock(kd_trace_mtx_sysctl);
 
        return(ret);
@@ -1236,69 +2624,66 @@ kdbg_control(int *name, __unused u_int namelen, user_addr_t where, size_t *sizep
 
 
 /*
- * This code can run concurrently with kernel_debug_internal()
- * without the need of any locks, because all reads of kd_bufptr[i],
- * which get modified by kernel_debug_internal(), are safe.
+ * This code can run for the most part concurrently with kernel_debug_internal()...
+ * 'release_storage_unit' will take the kds_spin_lock which may cause us to briefly
+ * synchronize with the recording side of this puzzle... otherwise, we are able to
+ * move through the lists w/o use of any locks
  */
 int
-kdbg_read(user_addr_t buffer, size_t *number)
+kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx)
 {
        unsigned int count;
-       unsigned int cpu;
-       int mincpu;
-       uint64_t  mintime, t, last_wrap_time;
-       int last_wrap_cpu;
+       unsigned int cpu, min_cpu;
+       uint64_t  mintime, t, barrier = 0;
        int error = 0;
        kd_buf *tempbuf;
+       uint32_t rcursor;
+       kd_buf lostevent;
+       union kds_ptr kdsp;
+       struct kd_storage *kdsp_actual;
+       struct kd_bufinfo *kdbp;
+       struct kd_bufinfo *min_kdbp;
        uint32_t tempbuf_count;
        uint32_t tempbuf_number;
-       unsigned int old_kdebug_flags, new_kdebug_flags;
-       unsigned int old_kdebug_slowcheck, new_kdebug_slowcheck;
+       uint32_t old_kdebug_flags;
+       uint32_t old_kdebug_slowcheck;
+       boolean_t lostevents = FALSE;
+       boolean_t out_of_events = FALSE;
+
        count = *number/sizeof(kd_buf);
        *number = 0;
 
-       if (count == 0 || !(kdebug_flags & KDBG_BUFINIT) || kdcopybuf == 0)
+       if (count == 0 || !(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) || kdcopybuf == 0)
                return EINVAL;
 
+       memset(&lostevent, 0, sizeof(lostevent));
+       lostevent.debugid = TRACE_LOST_EVENTS;
+
+       /* Capture timestamp. Only sort events that have occured before the timestamp.
+        * Since the iop is being flushed here, its possible that events occur on the AP
+        * while running live tracing. If we are disabled, no new events should 
+        * occur on the AP.
+       */
+       
+       if (kd_ctrl_page.enabled)
+       {
+               // timestamp is non-zero value
+               barrier = mach_absolute_time() & KDBG_TIMESTAMP_MASK;
+       }
+       
+       // Request each IOP to provide us with up to date entries before merging buffers together.
+       kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_SYNC_FLUSH, NULL);
+
        /*
         * because we hold kd_trace_mtx_sysctl, no other control threads can 
         * be playing with kdebug_flags... the code that cuts new events could
-        * be running, but it only reads kdebug_flags, it doesn't write it..
-        * use an OSCompareAndSwap to make sure the other processors see the 
-        * change of state immediately, not to protect against 2 threads racing to update it
+        * be running, but it grabs kds_spin_lock if it needs to acquire a new
+        * storage chunk which is where it examines kdebug_flags... it its adding
+        * to the same chunk we're reading from, no problem... 
         */
-       old_kdebug_slowcheck = kdebug_slowcheck;
-       do {
-               old_kdebug_flags = kdebug_flags;
-               new_kdebug_flags = old_kdebug_flags & ~KDBG_WRAPPED;
-               new_kdebug_flags |= KDBG_NOWRAP;
-       } while ( !OSCompareAndSwap((UInt32)old_kdebug_flags, (UInt32)new_kdebug_flags, (UInt32 *)&kdebug_flags));
-
-       last_wrap_time = 0;
-       last_wrap_cpu  = -1;
-
-       for (cpu = 0; cpu < kd_cpus; cpu++) {
-               kd_buf *cur_bufptr;
-               
-               if ((cur_bufptr = kdbip[cpu].kd_bufptr) >= kdbip[cpu].kd_buflast)
-                       cur_bufptr = kdbip[cpu].kd_buffer;
 
-               if (kdbip[cpu].kd_wrapped) {
-                       kdbip[cpu].kd_wrapped = 0;
-                       kdbip[cpu].kd_readlast = cur_bufptr;
-                       kdbip[cpu].kd_stop = cur_bufptr;
+       disable_wrap(&old_kdebug_slowcheck, &old_kdebug_flags);
 
-                       if (kd_cpus > 1 && ((cur_bufptr->timestamp & KDBG_TIMESTAMP_MASK) > last_wrap_time)) {
-                               last_wrap_time = cur_bufptr->timestamp & KDBG_TIMESTAMP_MASK;
-                               last_wrap_cpu = cpu;
-                       }
-               } else {
-                       if (kdbip[cpu].kd_readlast == cur_bufptr)
-                               kdbip[cpu].kd_stop = 0;
-                       else
-                               kdbip[cpu].kd_stop = cur_bufptr;
-               }
-       }
        if (count > nkdbufs)
                count = nkdbufs;
 
@@ -1306,67 +2691,145 @@ kdbg_read(user_addr_t buffer, size_t *number)
                tempbuf_count = KDCOPYBUF_COUNT;
 
        while (count) {
-               tempbuf = kdcopybuf;
+               tempbuf = kdcopybuf;
                tempbuf_number = 0;
 
-               while (tempbuf_count) {
-                       mintime = 0xffffffffffffffffULL; /* all actual timestamps are below */
-                       mincpu = -1;
+               // While space
+               while (tempbuf_count) {
+                       mintime = 0xffffffffffffffffULL;
+                       min_kdbp = NULL;
+                       min_cpu = 0;
 
-                       for (cpu = 0; cpu < kd_cpus; cpu++) {
-                               if (kdbip[cpu].kd_stop == 0)     /* empty buffer */
-                                       continue;
-                               t = kdbip[cpu].kd_readlast[0].timestamp & KDBG_TIMESTAMP_MASK;
+                       // Check all CPUs
+                       for (cpu = 0, kdbp = &kdbip[0]; cpu < kd_ctrl_page.kdebug_cpus; cpu++, kdbp++) {
 
+                               // Find one with raw data
+                               if ((kdsp = kdbp->kd_list_head).raw == KDS_PTR_NULL)
+                                       continue;
+                               /* Debugging aid: maintain a copy of the "kdsp"
+                                * index.
+                                */
+                               volatile union kds_ptr kdsp_shadow;
+
+                               kdsp_shadow = kdsp;
+
+                               // Get from cpu data to buffer header to buffer
+                               kdsp_actual = POINTER_FROM_KDS_PTR(kdsp);
+
+                               volatile struct kd_storage *kdsp_actual_shadow;
+
+                               kdsp_actual_shadow = kdsp_actual;
+
+                               // See if there are actual data left in this buffer
+                               rcursor = kdsp_actual->kds_readlast;
+
+                               if (rcursor == kdsp_actual->kds_bufindx)
+                                       continue;
+
+                               t = kdbg_get_timestamp(&kdsp_actual->kds_records[rcursor]);
+
+                               if ((t > barrier) && (barrier > 0)) {
+                                       /* 
+                                        * Need to wait to flush iop again before we 
+                                        * sort any more data from the buffers
+                                       */
+                                       out_of_events = TRUE;
+                                       break;
+                               }       
+                               if (t < kdsp_actual->kds_timestamp) {
+                                       /*
+                                        * indicates we've not yet completed filling
+                                        * in this event...
+                                        * this should only occur when we're looking
+                                        * at the buf that the record head is utilizing
+                                        * we'll pick these events up on the next
+                                        * call to kdbg_read
+                                        * we bail at this point so that we don't
+                                        * get an out-of-order timestream by continuing
+                                        * to read events from the other CPUs' timestream(s)
+                                        */
+                                       out_of_events = TRUE;
+                                       break;
+                               }
                                if (t < mintime) {
                                        mintime = t;
-                                       mincpu = cpu;
+                                       min_kdbp = kdbp;
+                                       min_cpu = cpu;
                                }
                        }
-                       if (mincpu < 0)
-                               /*
-                                * all buffers ran empty early
+                       if (min_kdbp == NULL || out_of_events == TRUE) {
+                               /*
+                                * all buffers ran empty
                                 */
-                               break;
-
-                       if (last_wrap_cpu == mincpu) {
-                               tempbuf->debugid = MISCDBG_CODE(DBG_BUFFER, 0) | DBG_FUNC_NONE;
-                               tempbuf->arg1 = 0;
-                               tempbuf->arg2 = 0;
-                               tempbuf->arg3 = 0;
-                               tempbuf->arg4 = 0;
-                               tempbuf->arg5 = (int)current_thread();
-                 
-                               tempbuf->timestamp = last_wrap_time | (((uint64_t)last_wrap_cpu) << KDBG_CPU_SHIFT);
+                               out_of_events = TRUE;
+                               break;
+                       }
 
-                               tempbuf++;
+                       // Get data
+                       kdsp = min_kdbp->kd_list_head;
+                       kdsp_actual = POINTER_FROM_KDS_PTR(kdsp);
 
-                               last_wrap_cpu = -1;
+                       if (kdsp_actual->kds_lostevents == TRUE) {
+                               kdbg_set_timestamp_and_cpu(&lostevent, kdsp_actual->kds_records[kdsp_actual->kds_readlast].timestamp, min_cpu);
+                               *tempbuf = lostevent;
+                               
+                               kdsp_actual->kds_lostevents = FALSE;
+                               lostevents = TRUE;
 
-                       } else {
-                               *(tempbuf++) = kdbip[mincpu].kd_readlast[0];
+                               goto nextevent;
+                       }
 
-                               kdbip[mincpu].kd_readlast++;
+                       // Copy into buffer
+                       *tempbuf = kdsp_actual->kds_records[kdsp_actual->kds_readlast++];
 
-                               if (kdbip[mincpu].kd_readlast == kdbip[mincpu].kd_buflast)
-                                       kdbip[mincpu].kd_readlast = kdbip[mincpu].kd_buffer;
-                               if (kdbip[mincpu].kd_readlast == kdbip[mincpu].kd_stop)
-                                       kdbip[mincpu].kd_stop = 0;
-                       }
+                       if (kdsp_actual->kds_readlast == EVENTS_PER_STORAGE_UNIT)
+                               release_storage_unit(min_cpu, kdsp.raw);
+
+                       /*
+                        * Watch for out of order timestamps
+                        */     
+                       if (mintime < min_kdbp->kd_prev_timebase) {
+                               /*
+                                * if so, use the previous timestamp + 1 cycle
+                                */
+                               min_kdbp->kd_prev_timebase++;
+                               kdbg_set_timestamp_and_cpu(tempbuf, min_kdbp->kd_prev_timebase, kdbg_get_cpu(tempbuf));
+                       } else
+                               min_kdbp->kd_prev_timebase = mintime;
+nextevent:
                        tempbuf_count--;
                        tempbuf_number++;
+                       tempbuf++;
+
+                       if ((RAW_file_written += sizeof(kd_buf)) >= RAW_FLUSH_SIZE)
+                               break;
                }
                if (tempbuf_number) {
-                       if ((error = copyout(kdcopybuf, buffer, tempbuf_number * sizeof(kd_buf)))) {
-                               *number = 0;
+
+                       if (vp) {
+                               error = vn_rdwr(UIO_WRITE, vp, (caddr_t)kdcopybuf, tempbuf_number * sizeof(kd_buf), RAW_file_offset,
+                                               UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
+
+                               RAW_file_offset += (tempbuf_number * sizeof(kd_buf));
+       
+                               if (RAW_file_written >= RAW_FLUSH_SIZE) {
+                                       cluster_push(vp, 0);
+
+                                       RAW_file_written = 0;
+                               }
+                       } else {
+                               error = copyout(kdcopybuf, buffer, tempbuf_number * sizeof(kd_buf));
+                               buffer += (tempbuf_number * sizeof(kd_buf));
+                       }
+                       if (error) {
+                               *number = 0;
                                error = EINVAL;
                                break;
                        }
                        count   -= tempbuf_number;
                        *number += tempbuf_number;
-                       buffer  += (tempbuf_number * sizeof(kd_buf));
                }
-               if (tempbuf_count)
+               if (out_of_events == TRUE)
                       /*
                        * all trace buffers are empty
                        */
@@ -1376,17 +2839,7 @@ kdbg_read(user_addr_t buffer, size_t *number)
                        tempbuf_count = KDCOPYBUF_COUNT;
        }
        if ( !(old_kdebug_flags & KDBG_NOWRAP)) {
-               do {
-                       old_kdebug_flags = kdebug_flags;
-                       new_kdebug_flags = old_kdebug_flags & ~KDBG_NOWRAP;
-               } while ( !OSCompareAndSwap((UInt32)old_kdebug_flags, (UInt32)new_kdebug_flags, (UInt32 *)&kdebug_flags));
-
-               if ( !(old_kdebug_slowcheck & SLOW_NOLOG)) {
-                       do {
-                               old_kdebug_slowcheck = kdebug_slowcheck;
-                               new_kdebug_slowcheck = old_kdebug_slowcheck & ~SLOW_NOLOG;
-                       } while ( !OSCompareAndSwap((UInt32)old_kdebug_slowcheck, (UInt32)new_kdebug_slowcheck, (UInt32 *)&kdebug_slowcheck));
-               }
+               enable_wrap(old_kdebug_slowcheck, lostevents);
        }
        return (error);
 }
@@ -1401,14 +2854,14 @@ unsigned char *getProcName(struct proc *proc) {
 
 #define STACKSHOT_SUBSYS_LOCK() lck_mtx_lock(&stackshot_subsys_mutex)
 #define STACKSHOT_SUBSYS_UNLOCK() lck_mtx_unlock(&stackshot_subsys_mutex)
-#ifdef __i386__
+#if defined(__i386__) || defined (__x86_64__)
 #define TRAP_DEBUGGER __asm__ volatile("int3");
-#endif
-#ifdef __ppc__
-#define TRAP_DEBUGGER   __asm__ volatile("tw 4,r3,r3");
+#else
+#error No TRAP_DEBUGGER definition for this architecture
 #endif
 
-#define SANE_TRACEBUF_SIZE 2*1024*1024
+#define SANE_TRACEBUF_SIZE (8 * 1024 * 1024)
+#define SANE_BOOTPROFILE_TRACEBUF_SIZE (64 * 1024 * 1024)
 
 /* Initialize the mutex governing access to the stack snapshot subsystem */
 __private_extern__ void
@@ -1446,45 +2899,183 @@ stackshot_lock_init( void )
  *                  tracebuffer exhaustion, we copyout as much as possible.
  */
 int
-stack_snapshot(struct proc *p, register struct stack_snapshot_args *uap, register_t *retval) {
+stack_snapshot(struct proc *p, register struct stack_snapshot_args *uap, int32_t *retval) {
        int error = 0;
 
        if ((error = suser(kauth_cred_get(), &p->p_acflag)))
                 return(error);
 
        return stack_snapshot2(uap->pid, uap->tracebuf, uap->tracebuf_size,
-           uap->options, retval);
+           uap->flags, uap->dispatch_offset, retval);
+}
+
+int  
+stack_snapshot_from_kernel(pid_t pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytesTraced)
+{
+       int error = 0;
+       boolean_t istate;
+
+       if ((buf == NULL) || (size <= 0) || (bytesTraced == NULL)) {
+               return -1;
+       }
+
+       /* cap in individual stackshot to SANE_TRACEBUF_SIZE */
+       if (size > SANE_TRACEBUF_SIZE) {
+               size = SANE_TRACEBUF_SIZE;
+       }
+
+/* Serialize tracing */        
+       STACKSHOT_SUBSYS_LOCK();
+       istate = ml_set_interrupts_enabled(FALSE);
+
+
+/* Preload trace parameters*/  
+       kdp_snapshot_preflight(pid, buf, size, flags, 0);
+
+/* Trap to the debugger to obtain a coherent stack snapshot; this populates
+ * the trace buffer
+ */
+       TRAP_DEBUGGER;
+
+       ml_set_interrupts_enabled(istate);
+
+       *bytesTraced = kdp_stack_snapshot_bytes_traced();
+
+       error = kdp_stack_snapshot_geterror();
+       
+       STACKSHOT_SUBSYS_UNLOCK();
+
+    return error;
+
 }
 
 int
-stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t options, register_t *retval)
+stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset, int32_t *retval)
 {
+       boolean_t istate;
        int error = 0;
        unsigned bytesTraced = 0;
 
+#if CONFIG_TELEMETRY
+       if (flags & STACKSHOT_GLOBAL_MICROSTACKSHOT_ENABLE) {
+               telemetry_global_ctl(1);
+               *retval = 0;
+               return (0);
+       } else if (flags & STACKSHOT_GLOBAL_MICROSTACKSHOT_DISABLE) {
+               telemetry_global_ctl(0);
+               *retval = 0;
+               return (0);
+       }
+
+       if (flags & STACKSHOT_WINDOWED_MICROSTACKSHOTS_ENABLE) {
+               error = telemetry_enable_window();
+
+               if (error != KERN_SUCCESS) {
+                       /* We are probably out of memory */
+                       *retval = -1;
+                       return ENOMEM;
+               }
+
+               *retval = 0;
+               return (0);
+       } else if (flags & STACKSHOT_WINDOWED_MICROSTACKSHOTS_DISABLE) {
+               telemetry_disable_window();
+               *retval = 0;
+               return (0);
+       }
+#endif
+
        *retval = -1;
 /* Serialize tracing */        
        STACKSHOT_SUBSYS_LOCK();
        
-       if ((tracebuf_size <= 0) || (tracebuf_size > SANE_TRACEBUF_SIZE)) {
+       if (tracebuf_size <= 0) {
+               error = EINVAL;
+               goto error_exit;
+       }
+
+#if CONFIG_TELEMETRY
+       if (flags & STACKSHOT_GET_MICROSTACKSHOT) {
+
+               if (tracebuf_size > SANE_TRACEBUF_SIZE) {
+                       error = EINVAL;
+                       goto error_exit;
+               }
+
+               bytesTraced = tracebuf_size;
+               error = telemetry_gather(tracebuf, &bytesTraced, 
+                                        (flags & STACKSHOT_SET_MICROSTACKSHOT_MARK) ? TRUE : FALSE);
+               if (error == KERN_NO_SPACE) {
+                       error = ENOSPC;
+               }
+
+               *retval = (int)bytesTraced;
+               goto error_exit;
+       }
+
+       if (flags & STACKSHOT_GET_WINDOWED_MICROSTACKSHOTS) {
+
+               if (tracebuf_size > SANE_TRACEBUF_SIZE) {
+                       error = EINVAL;
+                       goto error_exit;
+               }
+
+               bytesTraced = tracebuf_size;
+               error = telemetry_gather_windowed(tracebuf, &bytesTraced);
+               if (error == KERN_NO_SPACE) {
+                       error = ENOSPC;
+               }
+
+               *retval = (int)bytesTraced;
+               goto error_exit;
+       }
+
+       if (flags & STACKSHOT_GET_BOOT_PROFILE) {
+
+               if (tracebuf_size > SANE_BOOTPROFILE_TRACEBUF_SIZE) {
+                       error = EINVAL;
+                       goto error_exit;
+               }
+
+               bytesTraced = tracebuf_size;
+               error = bootprofile_gather(tracebuf, &bytesTraced);
+               if (error == KERN_NO_SPACE) {
+                       error = ENOSPC;
+               }
+
+               *retval = (int)bytesTraced;
+               goto error_exit;
+       }
+#endif
+
+       if (tracebuf_size > SANE_TRACEBUF_SIZE) {
                error = EINVAL;
                goto error_exit;
        }
 
-       MALLOC(stackshot_snapbuf, void *, tracebuf_size, M_TEMP, M_WAITOK);
+       assert(stackshot_snapbuf == NULL);
+       if (kmem_alloc_kobject(kernel_map, (vm_offset_t *)&stackshot_snapbuf, tracebuf_size) != KERN_SUCCESS) {
+               error = ENOMEM;
+               goto error_exit;
+       }
 
-       if (stackshot_snapbuf == NULL) {
+       if (panic_active()) {
                error = ENOMEM;
                goto error_exit;
        }
+
+       istate = ml_set_interrupts_enabled(FALSE);
 /* Preload trace parameters*/  
-       kdp_snapshot_preflight(pid, stackshot_snapbuf, tracebuf_size, options);
+       kdp_snapshot_preflight(pid, stackshot_snapbuf, tracebuf_size, flags, dispatch_offset);
 
 /* Trap to the debugger to obtain a coherent stack snapshot; this populates
  * the trace buffer
  */
+
        TRAP_DEBUGGER;
 
+       ml_set_interrupts_enabled(istate);
+
        bytesTraced = kdp_stack_snapshot_bytes_traced();
                        
        if (bytesTraced > 0) {
@@ -1508,8 +3099,288 @@ stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_
 
 error_exit:
        if (stackshot_snapbuf != NULL)
-               FREE(stackshot_snapbuf, M_TEMP);
+               kmem_free(kernel_map, (vm_offset_t) stackshot_snapbuf, tracebuf_size);
        stackshot_snapbuf = NULL;
        STACKSHOT_SUBSYS_UNLOCK();
        return error;
 }
+
+void
+start_kern_tracing(unsigned int new_nkdbufs, boolean_t need_map)
+{
+
+       if (!new_nkdbufs)
+               return;
+       nkdbufs = kdbg_set_nkdbufs(new_nkdbufs);
+       kdbg_lock_init();
+
+       kernel_debug_string("start_kern_tracing");
+
+       if (0 == kdbg_reinit(TRUE)) {
+
+               if (need_map == TRUE) {
+                       uint32_t old1, old2;
+
+                       kdbg_thrmap_init();
+
+                       disable_wrap(&old1, &old2);
+               }
+
+               /* Hold off interrupts until the early traces are cut */
+               boolean_t       s = ml_set_interrupts_enabled(FALSE);
+
+               kdbg_set_tracing_enabled(
+                       TRUE,
+                       kdebug_serial ?
+                               (KDEBUG_ENABLE_TRACE | KDEBUG_ENABLE_SERIAL) :
+                                KDEBUG_ENABLE_TRACE);
+
+               /*
+                * Transfer all very early events from the static buffer
+                * into the real buffers.
+                */
+               kernel_debug_early_end();
+       
+               ml_set_interrupts_enabled(s);
+
+               printf("kernel tracing started\n");
+#if KDEBUG_MOJO_TRACE
+               if (kdebug_serial) {
+                       printf("serial output enabled with %lu named events\n",
+                       sizeof(kd_events)/sizeof(kd_event_t));
+               }
+#endif
+       } else {
+               printf("error from kdbg_reinit, kernel tracing not started\n");
+       }
+}
+
+void
+start_kern_tracing_with_typefilter(unsigned int new_nkdbufs,
+                                  boolean_t need_map,
+                                  unsigned int typefilter)
+{
+       /* startup tracing */
+       start_kern_tracing(new_nkdbufs, need_map);
+
+       /* check that tracing was actually enabled */
+       if (!(kdebug_enable & KDEBUG_ENABLE_TRACE))
+               return;
+
+       /* setup the typefiltering */
+       if (0 == kdbg_enable_typefilter())
+               setbit(type_filter_bitmap, typefilter & (CSC_MASK >> CSC_OFFSET));
+}
+
+void
+kdbg_dump_trace_to_file(const char *filename)
+{
+       vfs_context_t   ctx;
+       vnode_t         vp;
+       int             error;
+       size_t          number;
+
+
+       if ( !(kdebug_enable & KDEBUG_ENABLE_TRACE))
+               return;
+
+        if (global_state_pid != -1) {
+               if ((proc_find(global_state_pid)) != NULL) {
+                       /*
+                        * The global pid exists, we're running
+                        * due to fs_usage, latency, etc...
+                        * don't cut the panic/shutdown trace file
+                        * Disable tracing from this point to avoid
+                        * perturbing state.
+                        */
+                       kdebug_enable = 0;
+                       kd_ctrl_page.enabled = 0;
+                       commpage_update_kdebug_enable();
+                       return;
+               }
+       }
+       KERNEL_DEBUG_CONSTANT(TRACE_PANIC | DBG_FUNC_NONE, 0, 0, 0, 0, 0);
+
+       kdebug_enable = 0;
+       kd_ctrl_page.enabled = 0;
+       commpage_update_kdebug_enable();
+
+       ctx = vfs_context_kernel();
+
+       if ((error = vnode_open(filename, (O_CREAT | FWRITE | O_NOFOLLOW), 0600, 0, &vp, ctx)))
+               return;
+
+       number = kd_mapcount * sizeof(kd_threadmap);
+       kdbg_readthrmap(0, &number, vp, ctx);
+
+       number = nkdbufs*sizeof(kd_buf);
+       kdbg_read(0, &number, vp, ctx);
+       
+       vnode_close(vp, FWRITE, ctx);
+
+       sync(current_proc(), (void *)NULL, (int *)NULL);
+}
+
+/* Helper function for filling in the BSD name for an address space
+ * Defined here because the machine bindings know only Mach threads
+ * and nothing about BSD processes.
+ *
+ * FIXME: need to grab a lock during this?
+ */
+void kdbg_get_task_name(char* name_buf, int len, task_t task)
+{
+       proc_t proc;
+       
+       /* Note: we can't use thread->task (and functions that rely on it) here 
+        * because it hasn't been initialized yet when this function is called.
+        * We use the explicitly-passed task parameter instead.
+        */
+       proc = get_bsdtask_info(task);
+       if (proc != PROC_NULL)
+               snprintf(name_buf, len, "%s/%d", proc->p_comm, proc->p_pid);
+       else
+               snprintf(name_buf, len, "%p [!bsd]", task);
+}
+
+#if KDEBUG_MOJO_TRACE
+static kd_event_t *
+binary_search(uint32_t id)
+{
+       int low, high, mid;
+
+       low = 0;
+       high = sizeof(kd_events)/sizeof(kd_event_t) - 1;
+
+       while (TRUE)
+       {
+               mid = (low + high) / 2;
+
+               if (low > high)
+                       return NULL; /* failed */
+               else if ( low + 1 >= high) {
+                       /* We have a match */
+                       if (kd_events[high].id == id)
+                               return &kd_events[high];
+                       else if (kd_events[low].id == id)
+                               return &kd_events[low];
+                       else
+                               return NULL;  /* search failed */
+               }
+               else if (id < kd_events[mid].id)
+                       high = mid;
+               else
+                       low = mid;
+       } 
+}
+
+/*
+ * Look up event id to get name string.
+ * Using a per-cpu cache of a single entry
+ * before resorting to a binary search of the full table.
+ */
+#define        NCACHE  1
+static kd_event_t      *last_hit[MAX_CPUS];
+static kd_event_t *
+event_lookup_cache(uint32_t cpu, uint32_t id)
+{
+       if (last_hit[cpu] == NULL || last_hit[cpu]->id != id)
+               last_hit[cpu] = binary_search(id);
+       return last_hit[cpu];
+}
+
+static uint64_t        kd_last_timstamp;
+
+static void
+kdebug_serial_print(
+       uint32_t        cpunum,
+       uint32_t        debugid,
+       uint64_t        timestamp,
+       uintptr_t       arg1,
+       uintptr_t       arg2,
+       uintptr_t       arg3,
+       uintptr_t       arg4,
+       uintptr_t       threadid
+       )
+{
+       char            kprintf_line[192];
+       char            event[40];
+       uint64_t        us = timestamp / NSEC_PER_USEC;
+       uint64_t        us_tenth = (timestamp % NSEC_PER_USEC) / 100;
+       uint64_t        delta = timestamp - kd_last_timstamp;
+       uint64_t        delta_us = delta / NSEC_PER_USEC;
+       uint64_t        delta_us_tenth = (delta % NSEC_PER_USEC) / 100;
+       uint32_t        event_id = debugid & DBG_FUNC_MASK;
+       const char      *command;
+       const char      *bra;
+       const char      *ket;
+       kd_event_t      *ep;
+
+       /* event time and delta from last */
+       snprintf(kprintf_line, sizeof(kprintf_line),
+               "%11llu.%1llu %8llu.%1llu ",
+               us, us_tenth, delta_us, delta_us_tenth);
+
+
+       /* event (id or name) - start prefixed by "[", end postfixed by "]" */
+       bra = (debugid & DBG_FUNC_START) ? "[" : " ";
+       ket = (debugid & DBG_FUNC_END)   ? "]" : " ";
+       ep = event_lookup_cache(cpunum, event_id);
+       if (ep) {
+               if (strlen(ep->name) < sizeof(event) - 3)
+                       snprintf(event, sizeof(event), "%s%s%s",
+                                bra, ep->name, ket);
+               else
+                       snprintf(event, sizeof(event), "%s%x(name too long)%s",
+                                bra, event_id, ket);
+       } else {
+               snprintf(event, sizeof(event), "%s%x%s",
+                        bra, event_id, ket);
+       }
+       snprintf(kprintf_line + strlen(kprintf_line),
+                sizeof(kprintf_line) - strlen(kprintf_line),
+                "%-40s  ", event);
+
+       /* arg1 .. arg4 with special cases for strings */
+       switch (event_id) {
+           case VFS_LOOKUP:
+           case VFS_LOOKUP_DONE:
+               if (debugid & DBG_FUNC_START) {
+                       /* arg1 hex then arg2..arg4 chars */
+                       snprintf(kprintf_line + strlen(kprintf_line),
+                               sizeof(kprintf_line) - strlen(kprintf_line),
+                               "%-16lx %-8s%-8s%-8s                          ",
+                               arg1, (char*)&arg2, (char*)&arg3, (char*)&arg4);
+                       break;
+               }
+               /* else fall through for arg1..arg4 chars */
+           case TRACE_STRING_EXEC:
+           case TRACE_STRING_NEWTHREAD:
+           case TRACE_INFO_STRING:
+               snprintf(kprintf_line + strlen(kprintf_line),
+                       sizeof(kprintf_line) - strlen(kprintf_line),
+                       "%-8s%-8s%-8s%-8s                                   ",
+                       (char*)&arg1, (char*)&arg2, (char*)&arg3, (char*)&arg4);
+               break;
+           default:
+               snprintf(kprintf_line + strlen(kprintf_line),
+                       sizeof(kprintf_line) - strlen(kprintf_line),
+                       "%-16lx %-16lx %-16lx %-16lx",
+                       arg1, arg2, arg3, arg4);
+       }
+
+       /* threadid, cpu and command name */
+       if (threadid == (uintptr_t)thread_tid(current_thread()) &&
+           current_proc() &&
+           current_proc()->p_comm)
+               command = current_proc()->p_comm;
+       else
+               command = "-";
+       snprintf(kprintf_line + strlen(kprintf_line),
+               sizeof(kprintf_line) - strlen(kprintf_line),
+               "  %-16lx  %-2d %s\n",
+               threadid, cpunum, command);
+       
+       kprintf("%s", kprintf_line);
+       kd_last_timstamp = timestamp;
+}
+#endif